Intel(R) 64 and IA-32 Architectures Software Developer`s Manual

Comments

Transcription

Intel(R) 64 and IA-32 Architectures Software Developer`s Manual
Intel® 64 and IA-32 Architectures
Software Developer’s Manual
Volume 3B:
System Programming Guide, Part 2
NOTE: The Intel® 64 and IA-32 Architectures Software Developer's Manual
consists of five volumes: Basic Architecture, Order Number 253665;
Instruction Set Reference A-M, Order Number 253666; Instruction Set
Reference N-Z, Order Number 253667; System Programming Guide,
Part 1, Order Number 253668; System Programming Guide, Part 2, Order
Number 253669. Refer to all five volumes when evaluating your design
needs.
Order Number: 253669-039US
May 2011
INFORMATION IN THIS DOCUMENT IS PROVIDED IN CONNECTION WITH INTEL PRODUCTS. NO LICENSE,
EXPRESS OR IMPLIED, BY ESTOPPEL OR OTHERWISE, TO ANY INTELLECTUAL PROPERTY RIGHTS IS GRANTED BY THIS DOCUMENT. EXCEPT AS PROVIDED IN INTEL'S TERMS AND CONDITIONS OF SALE FOR SUCH
PRODUCTS, INTEL ASSUMES NO LIABILITY WHATSOEVER AND INTEL DISCLAIMS ANY EXPRESS OR IMPLIED
WARRANTY, RELATING TO SALE AND/OR USE OF INTEL PRODUCTS INCLUDING LIABILITY OR WARRANTIES
RELATING TO FITNESS FOR A PARTICULAR PURPOSE, MERCHANTABILITY, OR INFRINGEMENT OF ANY
PATENT, COPYRIGHT OR OTHER INTELLECTUAL PROPERTY RIGHT.
UNLESS OTHERWISE AGREED IN WRITING BY INTEL, THE INTEL PRODUCTS ARE NOT DESIGNED NOR INTENDED FOR ANY APPLICATION IN WHICH THE FAILURE OF THE INTEL PRODUCT COULD CREATE A SITUATION WHERE PERSONAL INJURY OR DEATH MAY OCCUR.
Intel may make changes to specifications and product descriptions at any time, without notice. Designers
must not rely on the absence or characteristics of any features or instructions marked "reserved" or "undefined." Intel reserves these for future definition and shall have no responsibility whatsoever for conflicts or
incompatibilities arising from future changes to them. The information here is subject to change without notice. Do not finalize a design with this information.
The Intel® 64 architecture processors may contain design defects or errors known as errata. Current characterized errata are available on request.
Intel® Hyper-Threading Technology requires a computer system with an Intel® processor supporting Intel
Hyper-Threading Technology and an Intel® HT Technology enabled chipset, BIOS and operating system.
Performance will vary depending on the specific hardware and software you use. For more information, see
http://www.intel.com/technology/hyperthread/index.htm; including details on which processors support Intel HT
Technology.
Intel® Virtualization Technology requires a computer system with an enabled Intel® processor, BIOS, virtual
machine monitor (VMM) and for some uses, certain platform software enabled for it. Functionality, performance or other benefits will vary depending on hardware and software configurations. Intel® Virtualization
Technology-enabled BIOS and VMM applications are currently in development.
64-bit computing on Intel architecture requires a computer system with a processor, chipset, BIOS, operating system, device drivers and applications enabled for Intel® 64 architecture. Processors will not operate
(including 32-bit operation) without an Intel® 64 architecture-enabled BIOS. Performance will vary depending on your hardware and software configurations. Consult with your system vendor for more information.
Enabling Execute Disable Bit functionality requires a PC with a processor with Execute Disable Bit capability
and a supporting operating system. Check with your PC manufacturer on whether your system delivers Execute Disable Bit functionality.
Intel, Pentium, Intel Xeon, Intel NetBurst, Intel Core, Intel Core Solo, Intel Core Duo, Intel Core 2 Duo,
Intel Core 2 Extreme, Intel Pentium D, Itanium, Intel SpeedStep, MMX, Intel Atom, and VTune are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States and other countries.
*Other names and brands may be claimed as the property of others.
Contact your local Intel sales office or your distributor to obtain the latest specifications and before placing
your product order.
Copies of documents which have an ordering number and are referenced in this document, or other Intel
literature, may be obtained by calling 1-800-548-4725, or by visiting Intel’s website at http://www.intel.com
Copyright © 1997-2011 Intel Corporation
ii Vol. 3B
CHAPTER 20
INTRODUCTION TO VIRTUAL-MACHINE EXTENSIONS
20.1
OVERVIEW
This chapter describes the basics of virtual machine architecture and an overview of
the virtual-machine extensions (VMX) that support virtualization of processor hardware for multiple software environments.
Information about VMX instructions is provided in Intel® 64 and IA-32 Architectures
Software Developer’s Manual, Volume 2B. Other aspects of VMX and system
programming considerations are described in chapters of Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 3B.
20.2
VIRTUAL MACHINE ARCHITECTURE
Virtual-machine extensions define processor-level support for virtual machines on
IA-32 processors. Two principal classes of software are supported:
•
Virtual-machine monitors (VMM) — A VMM acts as a host and has full control
of the processor(s) and other platform hardware. A VMM presents guest software
(see next paragraph) with an abstraction of a virtual processor and allows it to
execute directly on a logical processor. A VMM is able to retain selective control of
processor resources, physical memory, interrupt management, and I/O.
•
Guest software — Each virtual machine (VM) is a guest software environment
that supports a stack consisting of operating system (OS) and application
software. Each operates independently of other virtual machines and uses on the
same interface to processor(s), memory, storage, graphics, and I/O provided by
a physical platform. The software stack acts as if it were running on a platform
with no VMM. Software executing in a virtual machine must operate with reduced
privilege so that the VMM can retain control of platform resources.
20.3
INTRODUCTION TO VMX OPERATION
Processor support for virtualization is provided by a form of processor operation
called VMX operation. There are two kinds of VMX operation: VMX root operation and
VMX non-root operation. In general, a VMM will run in VMX root operation and guest
software will run in VMX non-root operation. Transitions between VMX root operation
and VMX non-root operation are called VMX transitions. There are two kinds of VMX
transitions. Transitions into VMX non-root operation are called VM entries. Transitions from VMX non-root operation to VMX root operation are called VM exits.
Vol. 3B 20-1
INTRODUCTION TO VIRTUAL-MACHINE EXTENSIONS
Processor behavior in VMX root operation is very much as it is outside VMX operation.
The principal differences are that a set of new instructions (the VMX instructions) is
available and that the values that can be loaded into certain control registers are
limited (see Section 20.8).
Processor behavior in VMX non-root operation is restricted and modified to facilitate
virtualization. Instead of their ordinary operation, certain instructions (including the
new VMCALL instruction) and events cause VM exits to the VMM. Because these
VM exits replace ordinary behavior, the functionality of software in VMX non-root
operation is limited. It is this limitation that allows the VMM to retain control of
processor resources.
There is no software-visible bit whose setting indicates whether a logical processor is
in VMX non-root operation. This fact may allow a VMM to prevent guest software from
determining that it is running in a virtual machine.
Because VMX operation places restrictions even on software running with current
privilege level (CPL) 0, guest software can run at the privilege level for which it was
originally designed. This capability may simplify the development of a VMM.
20.4
LIFE CYCLE OF VMM SOFTWARE
Figure 20-1 illustrates the life cycle of a VMM and its guest software as well as the
interactions between them. The following items summarize that life cycle:
•
•
Software enters VMX operation by executing a VMXON instruction.
•
VM exits transfer control to an entry point specified by the VMM. The VMM can
take action appropriate to the cause of the VM exit and can then return to the
virtual machine using a VM entry.
•
Eventually, the VMM may decide to shut itself down and leave VMX operation. It
does so by executing the VMXOFF instruction.
Using VM entries, a VMM can then enter guests into virtual machines (one at a
time). The VMM effects a VM entry using instructions VMLAUNCH and
VMRESUME; it regains control using VM exits.
20-2 Vol. 3B
INTRODUCTION TO VIRTUAL-MACHINE EXTENSIONS
Guest 0
VM Exit
VMXON
Guest 1
VM Entry
VM Monitor
VM Exit
VMXOFF
Figure 20-1. Interaction of a Virtual-Machine Monitor and Guests
20.5
VIRTUAL-MACHINE CONTROL STRUCTURE
VMX non-root operation and VMX transitions are controlled by a data structure called
a virtual-machine control structure (VMCS).
Access to the VMCS is managed through a component of processor state called the
VMCS pointer (one per logical processor). The value of the VMCS pointer is the 64-bit
address of the VMCS. The VMCS pointer is read and written using the instructions
VMPTRST and VMPTRLD. The VMM configures a VMCS using the VMREAD, VMWRITE,
and VMCLEAR instructions.
A VMM could use a different VMCS for each virtual machine that it supports. For a
virtual machine with multiple logical processors (virtual processors), the VMM could
use a different VMCS for each virtual processor.
20.6
DISCOVERING SUPPORT FOR VMX
Before system software enters into VMX operation, it must discover the presence of
VMX support in the processor. System software can determine whether a processor
supports VMX operation using CPUID. If CPUID.1:ECX.VMX[bit 5] = 1, then VMX
operation is supported. See Chapter 3, “Instruction Set Reference, A-M” of Intel® 64
and IA-32 Architectures Software Developer’s Manual, Volume 2A.
The VMX architecture is designed to be extensible so that future processors in VMX
operation can support additional features not present in first-generation implementations of the VMX architecture. The availability of extensible VMX features is
reported to software using a set of VMX capability MSRs (see Appendix G, “VMX
Capability Reporting Facility”).
Vol. 3B 20-3
INTRODUCTION TO VIRTUAL-MACHINE EXTENSIONS
20.7
ENABLING AND ENTERING VMX OPERATION
Before system software can enter VMX operation, it enables VMX by setting
CR4.VMXE[bit 13] = 1. VMX operation is then entered by executing the VMXON
instruction. VMXON causes an invalid-opcode exception (#UD) if executed with
CR4.VMXE = 0. Once in VMX operation, it is not possible to clear CR4.VMXE (see
Section 20.8). System software leaves VMX operation by executing the VMXOFF
instruction. CR4.VMXE can be cleared outside of VMX operation after executing of
VMXOFF.
VMXON is also controlled by the IA32_FEATURE_CONTROL MSR (MSR address 3AH).
This MSR is cleared to zero when a logical processor is reset. The relevant bits of the
MSR are:
•
Bit 0 is the lock bit. If this bit is clear, VMXON causes a general-protection
exception. If the lock bit is set, WRMSR to this MSR causes a general-protection
exception; the MSR cannot be modified until a power-up reset condition. System
BIOS can use this bit to provide a setup option for BIOS to disable support for
VMX. To enable VMX support in a platform, BIOS must set bit 1, bit 2, or both
(see below), as well as the lock bit.
•
Bit 1 enables VMXON in SMX operation. If this bit is clear, execution of
VMXON in SMX operation causes a general-protection exception. Attempts to set
this bit on logical processors that do not support both VMX operation (see Section
20.6) and SMX operation (see Chapter 6, “Safer Mode Extensions Reference,” in
Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 2B)
cause general-protection exceptions.
•
Bit 2 enables VMXON outside SMX operation. If this bit is clear, execution of
VMXON outside SMX operation causes a general-protection exception. Attempts
to set this bit on logical processors that do not support VMX operation (see
Section 20.6) cause general-protection exceptions.
NOTE
A logical processor is in SMX operation if GETSEC[SEXIT] has not
been executed since the last execution of GETSEC[SENTER]. A logical
processor is outside SMX operation if GETSEC[SENTER] has not been
executed or if GETSEC[SEXIT] was executed after the last execution
of GETSEC[SENTER]. See Chapter 6, “Safer Mode Extensions
Reference,” in Intel® 64 and IA-32 Architectures Software
Developer’s Manual, Volume 2B.
Before executing VMXON, software should allocate a naturally aligned 4-KByte region
of memory that a logical processor may use to support VMX operation.1 This region
is called the VMXON region. The address of the VMXON region (the VMXON pointer)
1. Future processors may require that a different amount of memory be reserved. If so, this fact is
reported to software using the VMX capability-reporting mechanism.
20-4 Vol. 3B
INTRODUCTION TO VIRTUAL-MACHINE EXTENSIONS
is provided in an operand to VMXON. Section 21.10.5, “VMXON Region,” details how
software should initialize and access the VMXON region.
20.8
RESTRICTIONS ON VMX OPERATION
VMX operation places restrictions on processor operation. These are detailed below:
•
In VMX operation, processors may fix certain bits in CR0 and CR4 to specific
values and not support other values. VMXON fails if any of these bits contains an
unsupported value (see “VMXON—Enter VMX Operation” in Chapter 5 of the
Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 2B).
Any attempt to set one of these bits to an unsupported value while in VMX
operation (including VMX root operation) using any of the CLTS, LMSW, or MOV
CR instructions causes a general-protection exception. VM entry or VM exit
cannot set any of these bits to an unsupported value.2
NOTES
The first processors to support VMX operation require that the
following bits be 1 in VMX operation: CR0.PE, CR0.NE, CR0.PG, and
CR4.VMXE. The restrictions on CR0.PE and CR0.PG imply that VMX
operation is supported only in paged protected mode (including
IA-32e mode). Therefore, guest software cannot be run in unpaged
protected mode or in real-address mode. See Section 27.2,
“Supporting Processor Operating Modes in Guest Environments,” for
a discussion of how a VMM might support guest software that expects
to run in unpaged protected mode or in real-address mode.
Later processors support a VM-execution control called “unrestricted
guest” (see Section 21.6.2). If this control is 1, CR0.PE and CR0.PG
may be 0 in VMX non-root operation (even if the capability MSR
IA32_VMX_CR0_FIXED0 reports otherwise).3 Such processors allow
guest software to run in unpaged protected mode or in real-address
mode.
•
VMXON fails if a logical processor is in A20M mode (see “VMXON—Enter VMX
Operation” in Chapter 6 of the Intel® 64 and IA-32 Architectures Software
Developer’s Manual, Volume 2B). Once the processor is in VMX operation, A20M
2. Software should consult the VMX capability MSRs IA32_VMX_CR0_FIXED0 and
IA32_VMX_CR0_FIXED1 to determine how bits in CR0 are set. (see Appendix G.7). For CR4, software should consult the VMX capability MSRs IA32_VMX_CR4_FIXED0 and
IA32_VMX_CR4_FIXED1 (see Appendix G.8).
3. “Unrestricted guest” is a secondary processor-based VM-execution control. If bit 31 of the primary processor-based VM-execution controls is 0, VMX non-root operation functions as if the
“unrestricted guest” VM-execution control were 0. See Section 21.6.2.
Vol. 3B 20-5
INTRODUCTION TO VIRTUAL-MACHINE EXTENSIONS
interrupts are blocked. Thus, it is impossible to be in A20M mode in VMX
operation.
•
The INIT signal is blocked whenever a logical processor is in VMX root operation.
It is not blocked in VMX non-root operation. Instead, INITs cause VM exits (see
Section 22.3, “Other Causes of VM Exits”).
20-6 Vol. 3B
CHAPTER 21
VIRTUAL-MACHINE CONTROL STRUCTURES
21.1
OVERVIEW
A logical processor uses virtual-machine control data structures (VMCSs) while
it is in VMX operation. These manage transitions into and out of VMX non-root operation (VM entries and VM exits) as well as processor behavior in VMX non-root operation. This structure is manipulated by the new instructions VMCLEAR, VMPTRLD,
VMREAD, and VMWRITE.
A VMM can use a different VMCS for each virtual machine that it supports. For a
virtual machine with multiple logical processors (virtual processors), the VMM can
use a different VMCS for each virtual processor.
A logical processor associates a region in memory with each VMCS. This region is
called the VMCS region.1 Software references a specific VMCS using the 64-bit
physical address of the region (a VMCS pointer). VMCS pointers must be aligned on
a 4-KByte boundary (bits 11:0 must be zero). These pointers must not set bits
beyond the processor’s physical-address width.2,3
A logical processor may maintain a number of VMCSs that are active. The processor
may optimize VMX operation by maintaining the state of an active VMCS in memory,
on the processor, or both. At any given time, at most one of the active VMCSs is the
current VMCS. (This document frequently uses the term “the VMCS” to refer to the
current VMCS.) The VMLAUNCH, VMREAD, VMRESUME, and VMWRITE instructions
operate only on the current VMCS.
The following items describe how a logical processor determines which VMCSs are
active and which is current:
•
The memory operand of the VMPTRLD instruction is the address of a VMCS. After
execution of the instruction, that VMCS is both active and current on the logical
processor. Any other VMCS that had been active remains so, but no other VMCS
is current.
•
The memory operand of the VMCLEAR instruction is also the address of a VMCS.
After execution of the instruction, that VMCS is neither active nor current on the
1. The amount of memory required for a VMCS region is at most 4 KBytes. The exact size is implementation specific and can be determined by consulting the VMX capability MSR
IA32_VMX_BASIC to determine the size of the VMCS region (see Appendix G.1).
2. Software can determine a processor’s physical-address width by executing CPUID with
80000008H in EAX. The physical-address width is returned in bits 7:0 of EAX.
3. If IA32_VMX_BASIC[48] is read as 1, these pointers must not set any bits in the range 63:32; see
Appendix G.1.
Vol. 3B 21-1
VIRTUAL-MACHINE CONTROL STRUCTURES
logical processor. If the VMCS had been current on the logical processor, the
logical processor no longer has a current VMCS.
The VMPTRST instruction stores the address of the logical processor’s current VMCS
into a specified memory location (it stores the value FFFFFFFF_FFFFFFFFH if there is
no current VMCS).
The launch state of a VMCS determines which VM-entry instruction should be used
with that VMCS: the VMLAUNCH instruction requires a VMCS whose launch state is
“clear”; the VMRESUME instruction requires a VMCS whose launch state is
“launched”. A logical processor maintains a VMCS’s launch state in the corresponding
VMCS region. The following items describe how a logical processor manages the
launch state of a VMCS:
•
If the launch state of the current VMCS is “clear”, successful execution of the
VMLAUNCH instruction changes the launch state to “launched”.
•
The memory operand of the VMCLEAR instruction is the address of a VMCS. After
execution of the instruction, the launch state of that VMCS is “clear”.
•
There are no other ways to modify the launch state of a VMCS (it cannot be
modified using VMWRITE) and there is no direct way to discover it (it cannot be
read using VMREAD).
Figure 21-1 illustrates the different states of a VMCS. It uses “X” to refer to the VMCS
and “Y” to refer to any other VMCS. Thus: “VMPTRLD X” always makes X current and
active; “VMPTRLD Y” always makes X not current (because it makes Y current);
VMLAUNCH makes the launch state of X “launched” if X was current and its launch
state was “clear”; and VMCLEAR X always makes X inactive and not current and
makes its launch state “clear”.
The figure does not illustrate operations that do not modify the VMCS state relative
to these parameters (e.g., execution of VMPTRLD X when X is already current). Note
that VMCLEAR X makes X “inactive, not current, and clear,” even if X’s current state
is not defined (e.g., even if X has not yet been initialized). See Section 21.10.3.
21-2 Vol. 3B
VIRTUAL-MACHINE CONTROL STRUCTURES
VMCLEAR X
VMCLEAR X
X
VMLAUNCH
VMPTRLD Y
R
Anything
Else
Active
Not Current
Launched
VMPTRLD X
VMCLEAR X
A
LE
Active
Current
Clear
Inactive
Not Current
Clear
C
VM
VMPTRLD Y
VMPTRLD X
V
VM M P
CL TR
EA LD
R X
X
Active
Not Current
Clear
Active
Current
Launched
Figure 21-1. States of VMCS X
21.2
FORMAT OF THE VMCS REGION
A VMCS region comprises up to 4-KBytes.1 The format of a VMCS region is given in
Table 21-1.
Table 21-1. Format of the VMCS Region
Byte Offset
Contents
0
VMCS revision identifier
4
VMX-abort indicator
8
VMCS data (implementation-specific format)
The first 32 bits of the VMCS region contain the VMCS revision identifier. Processors that maintain VMCS data in different formats (see below) use different VMCS
1. The exact size is implementation specific and can be determined by consulting the VMX capability MSR IA32_VMX_BASIC to determine the size of the VMCS region (see Appendix G.1).
Vol. 3B 21-3
VIRTUAL-MACHINE CONTROL STRUCTURES
revision identifiers. These identifiers enable software to avoid using a VMCS region
formatted for one processor on a processor that uses a different format.1
Software should write the VMCS revision identifier to the VMCS region before using
that region for a VMCS. The VMCS revision identifier is never written by the
processor; VMPTRLD may fail if its operand references a VMCS region whose VMCS
revision identifier differs from that used by the processor. Software can discover the
VMCS revision identifier that a processor uses by reading the VMX capability MSR
IA32_VMX_BASIC (see Appendix G, “VMX Capability Reporting Facility”).
The next 32 bits of the VMCS region are used for the VMX-abort indicator. The
contents of these bits do not control processor operation in any way. A logical
processor writes a non-zero value into these bits if a VMX abort occurs (see Section
24.7). Software may also write into this field.
The remainder of the VMCS region is used for VMCS data (those parts of the VMCS
that control VMX non-root operation and the VMX transitions). The format of these
data is implementation-specific. VMCS data are discussed in Section 21.3 through
Section 21.9. To ensure proper behavior in VMX operation, software should maintain
the VMCS region and related structures (enumerated in Section 21.10.4) in
writeback cacheable memory. Future implementations may allow or require a
different memory type2. Software should consult the VMX capability MSR
IA32_VMX_BASIC (see Appendix G.1).
21.3
ORGANIZATION OF VMCS DATA
The VMCS data are organized into six logical groups:
•
Guest-state area. Processor state is saved into the guest-state area on
VM exits and loaded from there on VM entries.
•
•
Host-state area. Processor state is loaded from the host-state area on VM exits.
•
•
•
VM-exit control fields. These fields control VM exits.
VM-execution control fields. These fields control processor behavior in VMX
non-root operation. They determine in part the causes of VM exits.
VM-entry control fields. These fields control VM entries.
VM-exit information fields. These fields receive information on VM exits and
describe the cause and the nature of VM exits. They are read-only.
1. Logical processors that use the same VMCS revision identifier use the same size for VMCS
regions.
2. Alternatively, software may map any of these regions or structures with the UC memory type.
Doing so is strongly discouraged unless necessary as it will cause the performance of transitions
using those structures to suffer significantly. In addition, the processor will continue to use the
memory type reported in the VMX capability MSR IA32_VMX_BASIC with exceptions noted in
Appendix G.1.
21-4 Vol. 3B
VIRTUAL-MACHINE CONTROL STRUCTURES
The VM-execution control fields, the VM-exit control fields, and the VM-entry control
fields are sometimes referred to collectively as VMX controls.
21.4
GUEST-STATE AREA
This section describes fields contained in the guest-state area of the VMCS. As noted
earlier, processor state is loaded from these fields on every VM entry (see Section
23.3.2) and stored into these fields on every VM exit (see Section 24.3).
21.4.1
Guest Register State
The following fields in the guest-state area correspond to processor registers:
•
Control registers CR0, CR3, and CR4 (64 bits each; 32 bits on processors that do
not support Intel 64 architecture).
•
Debug register DR7 (64 bits; 32 bits on processors that do not support Intel 64
architecture).
•
RSP, RIP, and RFLAGS (64 bits each; 32 bits on processors that do not support
Intel 64 architecture).1
•
The following fields for each of the registers CS, SS, DS, ES, FS, GS, LDTR, and
TR:
— Selector (16 bits).
— Base address (64 bits; 32 bits on processors that do not support Intel 64
architecture). The base-address fields for CS, SS, DS, and ES have only 32
architecturally-defined bits; nevertheless, the corresponding VMCS fields
have 64 bits on processors that support Intel 64 architecture.
— Segment limit (32 bits). The limit field is always a measure in bytes.
— Access rights (32 bits). The format of this field is given in Table 21-2 and
detailed as follows:
•
The low 16 bits correspond to bits 23:8 of the upper 32 bits of a 64-bit
segment descriptor. While bits 19:16 of code-segment and data-segment
descriptors correspond to the upper 4 bits of the segment limit, the corresponding bits (bits 11:8) are reserved in this VMCS field.
1. This chapter uses the notation RAX, RIP, RSP, RFLAGS, etc. for processor registers because most
processors that support VMX operation also support Intel 64 architecture. For processors that do
not support Intel 64 architecture, this notation refers to the 32-bit forms of those registers
(EAX, EIP, ESP, EFLAGS, etc.). In a few places, notation such as EAX is used to refer specifically to
lower 32 bits of the indicated register.
Vol. 3B 21-5
VIRTUAL-MACHINE CONTROL STRUCTURES
•
Bit 16 indicates an unusable segment. Attempts to use such a segment
fault except in 64-bit mode. In general, a segment register is unusable if
it has been loaded with a null selector.1
•
Bits 31:17 are reserved.
Table 21-2. Format of Access Rights
Bit Position(s)
Field
3:0
Segment type
4
S — Descriptor type (0 = system; 1 = code or data)
6:5
DPL — Descriptor privilege level
7
P — Segment present
11:8
Reserved
12
AVL — Available for use by system software
13
Reserved (except for CS)
L — 64-bit mode active (for CS only)
14
D/B — Default operation size (0 = 16-bit segment; 1 = 32-bit segment)
15
G — Granularity
16
Segment unusable (0 = usable; 1 = unusable)
31:17
Reserved
The base address, segment limit, and access rights compose the “hidden” part
(or “descriptor cache”) of each segment register. These data are included in the
VMCS because it is possible for a segment register’s descriptor cache to be inconsistent with the segment descriptor in memory (in the GDT or the LDT)
referenced by the segment register’s selector.
The value of the DPL field for SS is always equal to the logical processor’s current
privilege level (CPL).2
•
The following fields for each of the registers GDTR and IDTR:
1. There are a few exceptions to this statement. For example, a segment with a non-null selector
may be unusable following a task switch that fails after its commit point; see “Interrupt
10—Invalid TSS Exception (#TS)” in Section 6.14, “Exception and Interrupt Handling in 64-bit
Mode,” of the Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 3A. In
contrast, the TR register is usable after processor reset despite having a null selector; see Table
10-1 in the Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 3A.
21-6 Vol. 3B
VIRTUAL-MACHINE CONTROL STRUCTURES
— Base address (64 bits; 32 bits on processors that do not support Intel 64
architecture).
— Limit (32 bits). The limit fields contain 32 bits even though these fields are
specified as only 16 bits in the architecture.
•
The following MSRs:
— IA32_DEBUGCTL (64 bits)
— IA32_SYSENTER_CS (32 bits)
— IA32_SYSENTER_ESP and IA32_SYSENTER_EIP (64 bits; 32 bits on
processors that do not support Intel 64 architecture)
— IA32_PERF_GLOBAL_CTRL (64 bits). This field is supported only on logical
processors that support the 1-setting of the “load IA32_PERF_GLOBAL_CTRL”
VM-entry control.
— IA32_PAT (64 bits). This field is supported only on logical processors that
support either the 1-setting of the “load IA32_PAT” VM-entry control or that
of the “save IA32_PAT” VM-exit control.
— IA32_EFER (64 bits). This field is supported only on logical processors that
support either the 1-setting of the “load IA32_EFER” VM-entry control or that
of the “save IA32_EFER” VM-exit control.
•
The register SMBASE (32 bits). This register contains the base address of the
logical processor’s SMRAM image.
21.4.2
Guest Non-Register State
In addition to the register state described in Section 21.4.1, the guest-state area
includes the following fields that characterize guest state but which do not correspond to processor registers:
•
Activity state (32 bits). This field identifies the logical processor’s activity state.
When a logical processor is executing instructions normally, it is in the active
state. Execution of certain instructions and the occurrence of certain events may
cause a logical processor to transition to an inactive state in which it ceases to
execute instructions.
The following activity states are defined:1
— 0: Active. The logical processor is executing instructions normally.
— 1: HLT. The logical processor is inactive because it executed the HLT
instruction.
2. In protected mode, CPL is also associated with the RPL field in the CS selector. However, the RPL
fields are not meaningful in real-address mode or in virtual-8086 mode.
1. Execution of the MWAIT instruction may put a logical processor into an inactive state. However,
this VMCS field never reflects this state. See Section 24.1.
Vol. 3B 21-7
VIRTUAL-MACHINE CONTROL STRUCTURES
— 2: Shutdown. The logical processor is inactive because it incurred a triple
fault1 or some other serious error.
— 3: Wait-for-SIPI. The logical processor is inactive because it is waiting for a
startup-IPI (SIPI).
Future processors may include support for other activity states. Software should
read the VMX capability MSR IA32_VMX_MISC (see Appendix G.6) to determine
what activity states are supported.
•
Interruptibility state (32 bits). The IA-32 architecture includes features that
permit certain events to be blocked for a period of time. This field contains
information about such blocking. Details and the format of this field are given in
Table 21-3.
Table 21-3. Format of Interruptibility State
Bit
Position(s)
Bit Name
Notes
0
Blocking by STI
See the “STI—Set Interrupt Flag” section in Chapter 4 of the
Intel® 64 and IA-32 Architectures Software Developer’s
Manual, Volume 2B.
Execution of STI with RFLAGS.IF = 0 blocks interrupts (and,
optionally, other events) for one instruction after its
execution. Setting this bit indicates that this blocking is in
effect.
1
Blocking by
MOV SS
See the “MOV—Move a Value from the Stack” and “POP—Pop
a Value from the Stack” sections in Chapter 3 and Chapter 4
of the Intel® 64 and IA-32 Architectures Software
Developer’s Manual, Volumes 2A & 2B, and Section 6.8.3 in
the Intel® 64 and IA-32 Architectures Software Developer’s
Manual, Volume 3A.
Execution of a MOV to SS or a POP to SS blocks interrupts for
one instruction after its execution. In addition, certain debug
exceptions are inhibited between a MOV to SS or a POP to SS
and a subsequent instruction. Setting this bit indicates that
the blocking of all these events is in effect. This document
uses the term “blocking by MOV SS,” but it applies equally to
POP SS.
2
Blocking by SMI
See Section 26.2. System-management interrupts (SMIs) are
disabled while the processor is in system-management mode
(SMM). Setting this bit indicates that blocking of SMIs is in
effect.
1. A triple fault occurs when a logical processor encounters an exception while attempting to
deliver a double fault.
21-8 Vol. 3B
VIRTUAL-MACHINE CONTROL STRUCTURES
Table 21-3. Format of Interruptibility State (Contd.)
Bit
Position(s)
Bit Name
Notes
3
Blocking by NMI
See Section 6.7.1 in the Intel® 64 and IA-32 Architectures
Software Developer’s Manual, Volume 3A and Section 26.8.
Delivery of a non-maskable interrupt (NMI) or a systemmanagement interrupt (SMI) blocks subsequent NMIs until
the next execution of IRET. See Section 22.4 for how this
behavior of IRET may change in VMX non-root operation.
Setting this bit indicates that blocking of NMIs is in effect.
Clearing this bit does not imply that NMIs are not
(temporarily) blocked for other reasons.
If the “virtual NMIs” VM-execution control (see Section
21.6.1) is 1, this bit does not control the blocking of NMIs.
Instead, it refers to “virtual-NMI blocking” (the fact that guest
software is not ready for an NMI).
31:4
•
Reserved
VM entry will fail if these bits are not 0. See Section 23.3.1.5.
Pending debug exceptions (64 bits; 32 bits on processors that do not support
Intel 64 architecture). IA-32 processors may recognize one or more debug
exceptions without immediately delivering them.1 This field contains information
about such exceptions. This field is described in Table 21-4.
Table 21-4. Format of Pending-Debug-Exceptions
Bit
Position(s)
Bit Name
Notes
3:0
B3 – B0
When set, each of these bits indicates that the corresponding
breakpoint condition was met. Any of these bits may be set
even if the corresponding enabling bit in DR7 is not set.
11:4
Reserved
VM entry fails if these bits are not 0. See Section 23.3.1.5.
12
Enabled
breakpoint
When set, this bit indicates that at least one data or I/O
breakpoint was met and was enabled in DR7.
1. For example, execution of a MOV to SS or a POP to SS may inhibit some debug exceptions for one
instruction. See Section 6.8.3 of Intel® 64 and IA-32 Architectures Software Developer’s Manual,
Volume 3A. In addition, certain events incident to an instruction (for example, an INIT signal) may
take priority over debug traps generated by that instruction. See Table 6-2 in the Intel® 64 and
IA-32 Architectures Software Developer’s Manual, Volume 3A.
Vol. 3B 21-9
VIRTUAL-MACHINE CONTROL STRUCTURES
Table 21-4. Format of Pending-Debug-Exceptions (Contd.)
Bit
Position(s)
Bit Name
Notes
13
Reserved
VM entry fails if this bit is not 0. See Section 23.3.1.5.
14
BS
When set, this bit indicates that a debug exception would
have been triggered by single-step execution mode.
63:15
Reserved
VM entry fails if these bits are not 0. See Section 23.3.1.5.
Bits 63:32 exist only on processors that support Intel 64
architecture.
•
VMCS link pointer (64 bits). This field is included for future expansion. Software
should set this field to FFFFFFFF_FFFFFFFFH to avoid VM-entry failures (see
Section 23.3.1.5).
•
VMX-preemption timer value (32 bits). This field is supported only on logical
processors that support the 1-setting of the “activate VMX-preemption timer”
VM-execution control. This field contains the value that the VMX-preemption
timer will use following the next VM entry with that setting. See Section 22.7.1
and Section 23.6.4.
•
Page-directory-pointer-table entries (PDPTEs; 64 bits each). These four (4)
fields (PDPTE0, PDPTE1, PDPTE2, and PDPTE3) are supported only on logical
processors that support the 1-setting of the “enable EPT” VM-execution control.
They correspond to the PDPTEs referenced by CR3 when PAE paging is in use (see
Section 4.4 in the Intel® 64 and IA-32 Architectures Software Developer’s
Manual, Volume 3A). They are used only if the “enable EPT” VM-execution control
is 1.
21.5
HOST-STATE AREA
This section describes fields contained in the host-state area of the VMCS. As noted
earlier, processor state is loaded from these fields on every VM exit (see Section
24.5).
All fields in the host-state area correspond to processor registers:
•
CR0, CR3, and CR4 (64 bits each; 32 bits on processors that do not support Intel
64 architecture).
•
RSP and RIP (64 bits each; 32 bits on processors that do not support Intel 64
architecture).
•
Selector fields (16 bits each) for the segment registers CS, SS, DS, ES, FS, GS,
and TR. There is no field in the host-state area for the LDTR selector.
•
Base-address fields for FS, GS, TR, GDTR, and IDTR (64 bits each; 32 bits on
processors that do not support Intel 64 architecture).
21-10 Vol. 3B
VIRTUAL-MACHINE CONTROL STRUCTURES
•
The following MSRs:
— IA32_SYSENTER_CS (32 bits)
— IA32_SYSENTER_ESP and IA32_SYSENTER_EIP (64 bits; 32 bits on
processors that do not support Intel 64 architecture).
— IA32_PERF_GLOBAL_CTRL (64 bits). This field is supported only on logical
processors that support the 1-setting of the “load IA32_PERF_GLOBAL_CTRL”
VM-exit control.
— IA32_PAT (64 bits). This field is supported only on logical processors that
support either the 1-setting of the “load IA32_PAT” VM-exit control.
— IA32_EFER (64 bits). This field is supported only on logical processors that
support either the 1-setting of the “load IA32_EFER” VM-exit control.
In addition to the state identified here, some processor state components are loaded
with fixed values on every VM exit; there are no fields corresponding to these components in the host-state area. See Section 24.5 for details of how state is loaded on
VM exits.
21.6
VM-EXECUTION CONTROL FIELDS
The VM-execution control fields govern VMX non-root operation. These are described
in Section 21.6.1 through Section 21.6.8.
21.6.1
Pin-Based VM-Execution Controls
The pin-based VM-execution controls constitute a 32-bit vector that governs the
handling of asynchronous events (for example: interrupts).1 Table 21-5 lists the
controls supported. See Chapter 22 for how these controls affect processor behavior
in VMX non-root operation.
1. Some asynchronous events cause VM exits regardless of the settings of the pin-based VM-execution controls (see Section 22.3).
Vol. 3B 21-11
VIRTUAL-MACHINE CONTROL STRUCTURES
Table 21-5. Definitions of Pin-Based VM-Execution Controls
Bit Position(s) Name
Description
0
External-interrupt
exiting
If this control is 1, external interrupts cause VM exits.
Otherwise, they are delivered normally through the guest
interrupt-descriptor table (IDT). If this control is 1, the value
of RFLAGS.IF does not affect interrupt blocking.
3
NMI exiting
If this control is 1, non-maskable interrupts (NMIs) cause
VM exits. Otherwise, they are delivered normally using
descriptor 2 of the IDT. This control also determines
interactions between IRET and blocking by NMI (see Section
22.4).
5
Virtual NMIs
If this control is 1, NMIs are never blocked and the “blocking
by NMI” bit (bit 3) in the interruptibility-state field indicates
“virtual-NMI blocking” (see Table 21-3). This control also
interacts with the “NMI-window exiting” VM-execution
control (see Section 21.6.2).
This control can be set only if the “NMI exiting” VM-execution
control (above) is 1.
6
Activate VMXpreemption timer
If this control is 1, the VMX-preemption timer counts down in
VMX non-root operation; see Section 22.7.1. A VM exit occurs
when the timer counts down to zero; see Section 22.3.
All other bits in this field are reserved, some to 0 and some to 1. Software should
consult the VMX capability MSRs IA32_VMX_PINBASED_CTLS and
IA32_VMX_TRUE_PINBASED_CTLS (see Appendix G.3.1) to determine how to set
reserved bits. Failure to set reserved bits properly causes subsequent VM entries to
fail (see Section 23.2).
The first processors to support the virtual-machine extensions supported only the 1settings of bits 1, 2, and 4. The VMX capability MSR IA32_VMX_PINBASED_CTLS will
always report that these bits must be 1. Logical processors that support the 0settings of any of these bits will support the VMX capability MSR
IA32_VMX_TRUE_PINBASED_CTLS MSR, and software should consult this MSR to
discover support for the 0-settings of these bits. Software that is not aware of the
functionality of any one of these bits should set that bit to 1.
21.6.2
Processor-Based VM-Execution Controls
The processor-based VM-execution controls constitute two 32-bit vectors that
govern the handling of synchronous events, mainly those caused by the execution of
specific instructions.1 These are the primary processor-based VM-execution
controls and the secondary processor-based VM-execution controls.
21-12 Vol. 3B
VIRTUAL-MACHINE CONTROL STRUCTURES
Table 21-6 lists the primary processor-based VM-execution controls. See Chapter 22
for more details of how these controls affect processor behavior in VMX non-root
operation.
Table 21-6. Definitions of Primary Processor-Based VM-Execution Controls
Bit Position(s) Name
Description
2
Interrupt-window
exiting
If this control is 1, a VM exit occurs at the beginning of any
instruction if RFLAGS.IF = 1 and there are no other blocking
of interrupts (see Section 21.4.2).
3
Use TSC offsetting
This control determines whether executions of RDTSC,
executions of RDTSCP, and executions of RDMSR that read
from the IA32_TIME_STAMP_COUNTER MSR return a value
modified by the TSC offset field (see Section 21.6.5 and
Section 22.4).
7
HLT exiting
This control determines whether executions of HLT cause
VM exits.
9
INVLPG exiting
This determines whether executions of INVLPG cause
VM exits.
10
MWAIT exiting
This control determines whether executions of MWAIT cause
VM exits.
11
RDPMC exiting
This control determines whether executions of RDPMC cause
VM exits.
12
RDTSC exiting
This control determines whether executions of RDTSC and
RDTSCP cause VM exits.
15
CR3-load exiting
In conjunction with the CR3-target controls (see Section
21.6.7), this control determines whether executions of MOV
to CR3 cause VM exits. See Section 22.1.3.
The first processors to support the virtual-machine
extensions supported only the 1-setting of this control.
16
CR3-store exiting
This control determines whether executions of MOV from
CR3 cause VM exits.
The first processors to support the virtual-machine
extensions supported only the 1-setting of this control.
19
CR8-load exiting
This control determines whether executions of MOV to CR8
cause VM exits.
This control must be 0 on processors that do not support
Intel 64 architecture.
1. Some instructions cause VM exits regardless of the settings of the processor-based VM-execution controls (see Section 22.1.2), as do task switches (see Section 22.3).
Vol. 3B 21-13
VIRTUAL-MACHINE CONTROL STRUCTURES
Table 21-6. Definitions of Primary Processor-Based VM-Execution Controls (Contd.)
Bit Position(s) Name
Description
20
This control determines whether executions of MOV from
CR8 cause VM exits.
CR8-store exiting
This control must be 0 on processors that do not support
Intel 64 architecture.
21
Use TPR shadow
Setting this control to 1 activates the TPR shadow, which is
maintained in a page of memory addressed by the virtualAPIC address. See Section 22.4.
This control must be 0 on processors that do not support
Intel 64 architecture.
22
NMI-window
exiting
If this control is 1, a VM exit occurs at the beginning of any
instruction if there is no virtual-NMI blocking (see Section
21.4.2).
This control can be set only if the “virtual NMIs” VMexecution control (see Section 21.6.1) is 1.
23
MOV-DR exiting
This control determines whether executions of MOV DR
cause VM exits.
24
Unconditional I/O
exiting
This control determines whether executions of I/O
instructions (IN, INS/INSB/INSW/INSD, OUT, and
OUTS/OUTSB/OUTSW/OUTSD) cause VM exits.
This control is ignored if the “use I/O bitmaps” control is 1.
25
Use I/O bitmaps
This control determines whether I/O bitmaps are used to
restrict executions of I/O instructions (see Section 21.6.4 and
Section 22.1.3).
For this control, “0” means “do not use I/O bitmaps” and “1”
means “use I/O bitmaps.” If the I/O bitmaps are used, the
setting of the “unconditional I/O exiting” control is ignored.
27
Monitor trap flag
If this control is 1, the monitor trap flag debugging feature is
enabled. See Section 22.7.2.
28
Use MSR bitmaps
This control determines whether MSR bitmaps are used to
control execution of the RDMSR and WRMSR instructions
(see Section 21.6.9 and Section 22.1.3).
For this control, “0” means “do not use MSR bitmaps” and “1”
means “use MSR bitmaps.” If the MSR bitmaps are not used,
all executions of the RDMSR and WRMSR instructions cause
VM exits.
29
21-14 Vol. 3B
MONITOR exiting
This control determines whether executions of MONITOR
cause VM exits.
VIRTUAL-MACHINE CONTROL STRUCTURES
Table 21-6. Definitions of Primary Processor-Based VM-Execution Controls (Contd.)
Bit Position(s) Name
Description
30
PAUSE exiting
This control determines whether executions of PAUSE cause
VM exits.
31
Activate secondary This control determines whether the secondary processorcontrols
based VM-execution controls are used. If this control is 0, the
logical processor operates as if all the secondary processorbased VM-execution controls were also 0.
All other bits in this field are reserved, some to 0 and some to 1. Software should
consult the VMX capability MSRs IA32_VMX_PROCBASED_CTLS and
IA32_VMX_TRUE_PROCBASED_CTLS (see Appendix G.3.2) to determine how to set
reserved bits. Failure to set reserved bits properly causes subsequent VM entries to
fail (see Section 23.2).
The first processors to support the virtual-machine extensions supported only the 1settings of bits 1, 4–6, 8, 13–16, and 26. The VMX capability MSR
IA32_VMX_PROCBASED_CTLS will always report that these bits must be 1. Logical
processors that support the 0-settings of any of these bits will support the VMX capability MSR IA32_VMX_TRUE_PROCBASED_CTLS MSR, and software should consult
this MSR to discover support for the 0-settings of these bits. Software that is not
aware of the functionality of any one of these bits should set that bit to 1.
Bit 31 of the primary processor-based VM-execution controls determines whether
the secondary processor-based VM-execution controls are used. If that bit is 0,
VM entry and VMX non-root operation function as if all the secondary processorbased VM-execution controls were 0. Processors that support only the 0-setting of
bit 31 of the primary processor-based VM-execution controls do not support the
secondary processor-based VM-execution controls.
Table 21-7 lists the secondary processor-based VM-execution controls. See Chapter
22 for more details of how these controls affect processor behavior in VMX non-root
operation.
Table 21-7. Definitions of Secondary Processor-Based VM-Execution Controls
Bit Position(s) Name
Description
0
Virtualize APIC
accesses
If this control is 1, a VM exit occurs on any attempt to access
data on the page with the APIC-access address. See Section
22.2.
1
Enable EPT
If this control is 1, extended page tables (EPT) are enabled.
See Section 25.2.
2
Descriptor-table
exiting
This control determines whether executions of LGDT, LIDT,
LLDT, LTR, SGDT, SIDT, SLDT, and STR cause VM exits.
3
Enable RDTSCP
If this control is 0, any execution of RDTSCP causes and
invalid-opcode exception (#UD).
Vol. 3B 21-15
VIRTUAL-MACHINE CONTROL STRUCTURES
Table 21-7. Definitions of Secondary Processor-Based VM-Execution Controls (Contd.)
Bit Position(s) Name
Description
4
Virtualize x2APIC
mode
Setting this control to 1 causes RDMSR and WRMSR to MSR
808H to use the TPR shadow, which is maintained on the
virtual-APIC page. See Section 22.4.
5
Enable VPID
If this control is 1, cached translations of linear addresses are
associated with a virtual-processor identifier (VPID). See
Section 25.1.
6
WBINVD exiting
This control determines whether executions of WBINVD
cause VM exits.
7
Unrestricted guest
This control determines whether guest software may run in
unpaged protected mode or in real-address mode.
10
PAUSE-loop exiting This control determines whether a series of executions of
PAUSE can cause a VM exit (see Section 21.6.13 and Section
22.1.3).
All other bits in these fields are reserved to 0. Software should consult the VMX capability MSR IA32_VMX_PROCBASED_CTLS2 (see Appendix G.3.3) to determine how to
set reserved bits. Failure to clear reserved bits causes subsequent VM entries to fail
(see Section 23.2).
If a logical processor supports the 1-setting of bit 31 of the primary processor-based
VM-execution controls but software has set that bit is 0, VM entry and VMX non-root
operation function as if all the secondary processor-based VM-execution controls
were 0. However, the logical processor will maintain the secondary processor-based
VM-execution controls as written by VMWRITE.
21.6.3
Exception Bitmap
The exception bitmap is a 32-bit field that contains one bit for each exception.
When an exception occurs, its vector is used to select a bit in this field. If the bit is 1,
the exception causes a VM exit. If the bit is 0, the exception is delivered normally
through the IDT, using the descriptor corresponding to the exception’s vector.
Whether a page fault (exception with vector 14) causes a VM exit is determined by
bit 14 in the exception bitmap as well as the error code produced by the page fault
and two 32-bit fields in the VMCS (the page-fault error-code mask and pagefault error-code match). See Section 22.3 for details.
21.6.4
I/O-Bitmap Addresses
The VM-execution control fields include the 64-bit physical addresses of I/O
bitmaps A and B (each of which are 4 KBytes in size). I/O bitmap A contains one bit
21-16 Vol. 3B
VIRTUAL-MACHINE CONTROL STRUCTURES
for each I/O port in the range 0000H through 7FFFH; I/O bitmap B contains bits for
ports in the range 8000H through FFFFH.
A logical processor uses these bitmaps if and only if the “use I/O bitmaps” control is
1. If the bitmaps are used, execution of an I/O instruction causes a VM exit if any bit
in the I/O bitmaps corresponding to a port it accesses is 1. See Section 22.1.3 for
details. If the bitmaps are used, their addresses must be 4-KByte aligned.
21.6.5
Time-Stamp Counter Offset
VM-execution control fields include a 64-bit TSC-offset field. If the “RDTSC exiting”
control is 0 and the “use TSC offsetting” control is 1, this field controls executions of
the RDTSC and RDTSCP instructions. It also controls executions of the RDMSR
instruction that read from the IA32_TIME_STAMP_COUNTER MSR. For all of these,
the signed value of the TSC offset is combined with the contents of the time-stamp
counter (using signed addition) and the sum is reported to guest software in
EDX:EAX. See Chapter 22 for a detailed treatment of the behavior of RDTSC,
RDTSCP, and RDMSR in VMX non-root operation.
21.6.6
Guest/Host Masks and Read Shadows for CR0 and CR4
VM-execution control fields include guest/host masks and read shadows for the
CR0 and CR4 registers. These fields control executions of instructions that access
those registers (including CLTS, LMSW, MOV CR, and SMSW). They are 64 bits on
processors that support Intel 64 architecture and 32 bits on processors that do not.
In general, bits set to 1 in a guest/host mask correspond to bits “owned” by the host:
•
Guest attempts to set them (using CLTS, LMSW, or MOV to CR) to values differing
from the corresponding bits in the corresponding read shadow cause VM exits.
•
Guest reads (using MOV from CR or SMSW) return values for these bits from the
corresponding read shadow.
Bits cleared to 0 correspond to bits “owned” by the guest; guest attempts to modify
them succeed and guest reads return values for these bits from the control register
itself.
See Chapter 22 for details regarding how these fields affect VMX non-root operation.
21.6.7
CR3-Target Controls
The VM-execution control fields include a set of 4 CR3-target values and a CR3target count. The CR3-target values each have 64 bits on processors that support
Intel 64 architecture and 32 bits on processors that do not. The CR3-target count has
32 bits on all processors.
An execution of MOV to CR3 in VMX non-root operation does not cause a VM exit if its
source operand matches one of these values. If the CR3-target count is n, only the
Vol. 3B 21-17
VIRTUAL-MACHINE CONTROL STRUCTURES
first n CR3-target values are considered; if the CR3-target count is 0, MOV to CR3
always causes a VM exit
There are no limitations on the values that can be written for the CR3-target values.
VM entry fails (see Section 23.2) if the CR3-target count is greater than 4.
Future processors may support a different number of CR3-target values. Software
should read the VMX capability MSR IA32_VMX_MISC (see Appendix G.6) to determine the number of values supported.
21.6.8
Controls for APIC Accesses
There are three mechanisms by which software accesses registers of the logical
processor’s local APIC:
•
If the local APIC is in xAPIC mode, it can perform memory-mapped accesses to
addresses in the 4-KByte page referenced by the physical address in the
IA32_APIC_BASE MSR (see Section 10.4.4, “Local APIC Status and Location” in
the Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 3A
and Intel® 64 Architecture Processor Topology Enumeration).1
•
If the local APIC is in x2APIC mode, it can accesses the local APIC’s registers
using the RDMSR and WRMSR instructions (see Intel® 64 Architecture Processor
Topology Enumeration).
•
In 64-bit mode, it can access the local APIC’s task-priority register (TPR) using
the MOV CR8 instruction.
There are three processor-based VM-execution controls (see Section 21.6.2) that
control such accesses. There are “use TPR shadow”, “virtualize APIC accesses”, and
“virtualize x2APIC mode”. These controls interact with the following fields:
•
APIC-access address (64 bits). This field is the physical address of the 4-KByte
APIC-access page. If the “virtualize APIC accesses” VM-execution control is 1,
operations that access this page may cause VM exits. See Section 22.2 and
Section 22.5.
The APIC-access address exists only on processors that support the 1-setting of
the “virtualize APIC accesses” VM-execution control.
•
Virtual-APIC address (64 bits). This field is the physical address of the 4-KByte
virtual-APIC page.
If the “use TPR shadow” VM-execution control is 1, the virtual-APIC address must
be 4-KByte aligned. The virtual-APIC page is accessed by the following
operations if the “use TPR shadow” VM-execution control is 1:
— The MOV CR8 instructions (see Section 22.1.3 and Section 22.4).
— Accesses to byte 80H on the APIC-access page if, in addition, the “virtualize
APIC accesses” VM-execution control is 1 (see Section 22.5.3).
1. If the local APIC does not support x2APIC mode, it is always in xAPIC mode.
21-18 Vol. 3B
VIRTUAL-MACHINE CONTROL STRUCTURES
— The RDMSR and WRMSR instructions if, in addition, the value of ECX is 808H
(indicating the TPR MSR) and the “virtualize x2APIC mode” VM-execution
control is 1 (see Section 22.4).
The virtual-APIC address exists only on processors that support the 1-setting of
the “use TPR shadow” VM-execution control.
•
TPR threshold (32 bits). Bits 3:0 of this field determine the threshold below
which the TPR shadow (bits 7:4 of byte 80H of the virtual-APIC page) cannot fall.
A VM exit occurs after an operation (e.g., an execution of MOV to CR8) that
reduces the TPR shadow below this value. See Section 22.4 and Section 22.5.3.
The TPR threshold exists only on processors that support the 1-setting of the
“use TPR shadow” VM-execution control.
21.6.9
MSR-Bitmap Address
On processors that support the 1-setting of the “use MSR bitmaps” VM-execution
control, the VM-execution control fields include the 64-bit physical address of four
contiguous MSR bitmaps, which are each 1-KByte in size. This field does not exist
on processors that do not support the 1-setting of that control. The four bitmaps are:
•
Read bitmap for low MSRs (located at the MSR-bitmap address). This contains
one bit for each MSR address in the range 00000000H to 00001FFFH. The bit
determines whether an execution of RDMSR applied to that MSR causes a
VM exit.
•
Read bitmap for high MSRs (located at the MSR-bitmap address plus 1024).
This contains one bit for each MSR address in the range C0000000H
toC0001FFFH. The bit determines whether an execution of RDMSR applied to that
MSR causes a VM exit.
•
Write bitmap for low MSRs (located at the MSR-bitmap address plus 2048).
This contains one bit for each MSR address in the range 00000000H to
00001FFFH. The bit determines whether an execution of WRMSR applied to that
MSR causes a VM exit.
•
Write bitmap for high MSRs (located at the MSR-bitmap address plus 3072).
This contains one bit for each MSR address in the range C0000000H
toC0001FFFH. The bit determines whether an execution of WRMSR applied to
that MSR causes a VM exit.
A logical processor uses these bitmaps if and only if the “use MSR bitmaps” control
is 1. If the bitmaps are used, an execution of RDMSR or WRMSR causes a VM exit if
the value of RCX is in neither of the ranges covered by the bitmaps or if the appropriate bit in the MSR bitmaps (corresponding to the instruction and the RCX value) is
1. See Section 22.1.3 for details. If the bitmaps are used, their address must be 4KByte aligned.
Vol. 3B 21-19
VIRTUAL-MACHINE CONTROL STRUCTURES
21.6.10 Executive-VMCS Pointer
The executive-VMCS pointer is a 64-bit field used in the dual-monitor treatment of
system-management interrupts (SMIs) and system-management mode (SMM). SMM
VM exits save this field as described in Section 26.15.2. VM entries that return from
SMM use this field as described in Section 26.15.4.
21.6.11 Extended-Page-Table Pointer (EPTP)
The extended-page-table pointer (EPTP) contains the address of the base of EPT
PML4 table (see Section 25.2.2), as well as other EPT configuration information. The
format of this field is shown in Table 21-8.
Table 21-8. Format of Extended-Page-Table Pointer
Bit Position(s)
Field
2:0
EPT paging-structure memory type (see Section 25.2.4):
0 = Uncacheable (UC)
6 = Write-back (WB)
Other values are reserved.1
5:3
This value is 1 less than the EPT page-walk length (see Section 25.2.2)
11:6
Reserved
N–1:12
Bits N–1:12 of the physical address of the 4-KByte aligned EPT PML4 table2
63:N
Reserved
NOTES:
1. Software should read the VMX capability MSR IA32_VMX_EPT_VPID_CAP (see Appendix G.10) to
determine what EPT paging-structure memory types are supported.
2. N is the physical-address width supported by the logical processor. Software can determine a processor’s physical-address width by executing CPUID with 80000008H in EAX. The physicaladdress width is returned in bits 7:0 of EAX.
The EPTP exists only on processors that support the 1-setting of the “enable EPT”
VM-execution control.
21.6.12 Virtual-Processor Identifier (VPID)
The virtual-processor identifier (VPID) is a 16-bit field. It exists only on processors that support the 1-setting of the “enable VPID” VM-execution control. See
Section 25.1 for details regarding the use of this field.
21-20 Vol. 3B
VIRTUAL-MACHINE CONTROL STRUCTURES
21.6.13 Controls for PAUSE-Loop Exiting
On processors that support the 1-setting of the “PAUSE-loop exiting” VM-execution
control, the VM-execution control fields include the following 32-bit fields:
•
PLE_Gap. Software can configure this field as an upper bound on the amount of
time between two successive executions of PAUSE in a loop.
•
PLE_Window. Software can configure this field as an upper bound on the
amount of time a guest is allowed to execute in a PAUSE loop.
These fields measure time based on a counter that runs at the same rate as the
timestamp counter (TSC). See Section 22.1.3 for more details regarding PAUSE-loop
exiting.
21.7
VM-EXIT CONTROL FIELDS
The VM-exit control fields govern the behavior of VM exits. They are discussed in
Section 21.7.1 and Section 21.7.2.
21.7.1
VM-Exit Controls
The VM-exit controls constitute a 32-bit vector that governs the basic operation of
VM exits. Table 21-9 lists the controls supported. See Chapter 24 for complete details
of how these controls affect VM exits.
Table 21-9. Definitions of VM-Exit Controls
Bit Position(s) Name
Description
2
This control determines whether DR7 and the
IA32_DEBUGCTL MSR are saved on VM exit.
Save debug
controls
The first processors to support the virtual-machine
extensions supported only the 1-setting of this control.
9
Host addressspace size
On processors that support Intel 64 architecture, this
control determines whether a logical processor is in 64-bit
mode after the next VM exit. Its value is loaded into CS.L,
IA32_EFER.LME, and IA32_EFER.LMA on every VM exit.1
This control must be 0 on processors that do not support
Intel 64 architecture.
12
Load
IA32_PERF_GLOB
AL_CTRL
This control determines whether the
IA32_PERF_GLOBAL_CTRL MSR is loaded on VM exit.
Vol. 3B 21-21
VIRTUAL-MACHINE CONTROL STRUCTURES
Table 21-9. Definitions of VM-Exit Controls (Contd.)
Bit Position(s) Name
Description
15
This control affects VM exits due to external interrupts:
Acknowledge
interrupt on exit
• If such a VM exit occurs and this control is 1, the logical
processor acknowledges the interrupt controller,
acquiring the interrupt’s vector. The vector is stored in
the VM-exit interruption-information field, which is
marked valid.
• If such a VM exit occurs and this control is 0, the
interrupt is not acknowledged and the VM-exit
interruption-information field is marked invalid.
18
Save IA32_PAT
This control determines whether the IA32_PAT MSR is
saved on VM exit.
19
Load IA32_PAT
This control determines whether the IA32_PAT MSR is
loaded on VM exit.
20
Save IA32_EFER
This control determines whether the IA32_EFER MSR is
saved on VM exit.
21
Load IA32_EFER
This control determines whether the IA32_EFER MSR is
loaded on VM exit.
22
Save VMXpreemption timer
value
This control determines whether the value of the VMXpreemption timer is saved on VM exit.
NOTES:
1. Since Intel 64 architecture specifies that IA32_EFER.LMA is always set to the logical-AND of
CR0.PG and IA32_EFER.LME, and since CR0.PG is always 1 in VMX operation, IA32_EFER.LMA is
always identical to IA32_EFER.LME in VMX operation.
All other bits in this field are reserved, some to 0 and some to 1. Software should
consult the VMX capability MSRs IA32_VMX_EXIT_CTLS and
IA32_VMX_TRUE_EXIT_CTLS (see Appendix G.4) to determine how it should set the
reserved bits. Failure to set reserved bits properly causes subsequent VM entries to
fail (see Section 23.2).
The first processors to support the virtual-machine extensions supported only the 1settings of bits 0–8, 10, 11, 13, 14, 16, and 17. The VMX capability MSR
IA32_VMX_EXIT_CTLS always reports that these bits must be 1. Logical processors
that support the 0-settings of any of these bits will support the VMX capability MSR
IA32_VMX_TRUE_EXIT_CTLS MSR, and software should consult this MSR to discover
support for the 0-settings of these bits. Software that is not aware of the functionality
of any one of these bits should set that bit to 1.
21-22 Vol. 3B
VIRTUAL-MACHINE CONTROL STRUCTURES
21.7.2
VM-Exit Controls for MSRs
A VMM may specify lists of MSRs to be stored and loaded on VM exits. The following
VM-exit control fields determine how MSRs are stored on VM exits:
•
VM-exit MSR-store count (32 bits). This field specifies the number of MSRs to
be stored on VM exit. It is recommended that this count not exceed 512 bytes.1
Otherwise, unpredictable processor behavior (including a machine check) may
result during VM exit.
•
VM-exit MSR-store address (64 bits). This field contains the physical address
of the VM-exit MSR-store area. The area is a table of entries, 16 bytes per entry,
where the number of entries is given by the VM-exit MSR-store count. The format
of each entry is given in Table 21-10. If the VM-exit MSR-store count is not zero,
the address must be 16-byte aligned.
Table 21-10. Format of an MSR Entry
Bit Position(s) Contents
31:0
MSR index
63:32
Reserved
127:64
MSR data
See Section 24.4 for how this area is used on VM exits.
The following VM-exit control fields determine how MSRs are loaded on VM exits:
•
VM-exit MSR-load count (32 bits). This field contains the number of MSRs to
be loaded on VM exit. It is recommended that this count not exceed 512 bytes.
Otherwise, unpredictable processor behavior (including a machine check) may
result during VM exit.2
•
VM-exit MSR-load address (64 bits). This field contains the physical address of
the VM-exit MSR-load area. The area is a table of entries, 16 bytes per entry,
where the number of entries is given by the VM-exit MSR-load count (see
Table 21-10). If the VM-exit MSR-load count is not zero, the address must be
16-byte aligned.
See Section 24.6 for how this area is used on VM exits.
1. Future implementations may allow more MSRs to be stored reliably. Software should consult the
VMX capability MSR IA32_VMX_MISC to determine the number supported (see Appendix G.6).
2. Future implementations may allow more MSRs to be loaded reliably. Software should consult the
VMX capability MSR IA32_VMX_MISC to determine the number supported (see Appendix G.6).
Vol. 3B 21-23
VIRTUAL-MACHINE CONTROL STRUCTURES
21.8
VM-ENTRY CONTROL FIELDS
The VM-entry control fields govern the behavior of VM entries. They are discussed in
Sections 21.8.1 through 21.8.3.
21.8.1
VM-Entry Controls
The VM-entry controls constitute a 32-bit vector that governs the basic operation of
VM entries. Table 21-11 lists the controls supported. See Chapter 23 for how these
controls affect VM entries.
Table 21-11. Definitions of VM-Entry Controls
Bit Position(s) Name
Description
2
This control determines whether DR7 and the
IA32_DEBUGCTL MSR are loaded on VM exit.
Load debug
controls
The first processors to support the virtual-machine
extensions supported only the 1-setting of this control.
9
IA-32e mode guest On processors that support Intel 64 architecture, this control
determines whether the logical processor is in IA-32e mode
after VM entry. Its value is loaded into IA32_EFER.LMA as
part of VM entry.1
This control must be 0 on processors that do not support
Intel 64 architecture.
10
Entry to SMM
This control determines whether the logical processor is in
system-management mode (SMM) after VM entry. This
control must be 0 for any VM entry from outside SMM.
11
Deactivate dualmonitor treatment
If set to 1, the default treatment of SMIs and SMM is in effect
after the VM entry (see Section 26.15.7). This control must
be 0 for any VM entry from outside SMM.
13
Load
This control determines whether the
IA32_PERF_GLOBA IA32_PERF_GLOBAL_CTRL MSR is loaded on VM entry.
L_CTRL
14
Load IA32_PAT
This control determines whether the IA32_PAT MSR is
loaded on VM entry.
15
Load IA32_EFER
This control determines whether the IA32_EFER MSR is
loaded on VM entry.
NOTES:
1. Bit 5 of the IA32_VMX_MISC MSR is read as 1 on any logical processor that supports the 1-setting
of the “unrestricted guest” VM-execution control. If it is read as 1, every VM exit stores the value of
IA32_EFER.LMA into the “IA-32e mode guest” VM-entry control (see Section 24.2).
21-24 Vol. 3B
VIRTUAL-MACHINE CONTROL STRUCTURES
All other bits in this field are reserved, some to 0 and some to 1. Software should
consult the VMX capability MSRs IA32_VMX_ENTRY_CTLS and
IA32_VMX_TRUE_ENTRY_CTLS (see Appendix G.5) to determine how it should set
the reserved bits. Failure to set reserved bits properly causes subsequent VM entries
to fail (see Section 23.2).
The first processors to support the virtual-machine extensions supported only the 1settings of bits 0–8 and 12. The VMX capability MSR IA32_VMX_ENTRY_CTLS always
reports that these bits must be 1. Logical processors that support the 0-settings of
any of these bits will support the VMX capability MSR IA32_VMX_TRUE_ENTRY_CTLS
MSR, and software should consult this MSR to discover support for the 0-settings of
these bits. Software that is not aware of the functionality of any one of these bits
should set that bit to 1.
21.8.2
VM-Entry Controls for MSRs
A VMM may specify a list of MSRs to be loaded on VM entries. The following VM-entry
control fields manage this functionality:
•
VM-entry MSR-load count (32 bits). This field contains the number of MSRs to
be loaded on VM entry. It is recommended that this count not exceed 512 bytes.
Otherwise, unpredictable processor behavior (including a machine check) may
result during VM entry.1
•
VM-entry MSR-load address (64 bits). This field contains the physical address
of the VM-entry MSR-load area. The area is a table of entries, 16 bytes per entry,
where the number of entries is given by the VM-entry MSR-load count. The
format of entries is described in Table 21-10. If the VM-entry MSR-load count is
not zero, the address must be 16-byte aligned.
See Section 23.4 for details of how this area is used on VM entries.
21.8.3
VM-Entry Controls for Event Injection
VM entry can be configured to conclude by delivering an event through the IDT (after
all guest state and MSRs have been loaded). This process is called event injection
and is controlled by the following three VM-entry control fields:
•
VM-entry interruption-information field (32 bits). This field provides details
about the event to be injected. Table 21-12 describes the field.
Table 21-12. Format of the VM-Entry Interruption-Information Field
Bit
Position(s)
Content
7:0
Vector of interrupt or exception
1. Future implementations may allow more MSRs to be loaded reliably. Software should consult the
VMX capability MSR IA32_VMX_MISC to determine the number supported (see Appendix G.6).
Vol. 3B 21-25
VIRTUAL-MACHINE CONTROL STRUCTURES
Table 21-12. Format of the VM-Entry Interruption-Information Field (Contd.)
Bit
Position(s)
Content
10:8
Interruption type:
0: External interrupt
1: Reserved
2: Non-maskable interrupt (NMI)
3: Hardware exception
4: Software interrupt
5: Privileged software exception
6: Software exception
7: Other event
11
Deliver error code (0 = do not deliver; 1 = deliver)
30:12
Reserved
31
Valid
— The vector (bits 7:0) determines which entry in the IDT is used or which
other event is injected.
— The interruption type (bits 10:8) determines details of how the injection is
performed. In general, a VMM should use the type hardware exception for
all exceptions other than breakpoint exceptions (#BP; generated by INT3)
and overflow exceptions (#OF; generated by INTO); it should use the type
software exception for #BP and #OF. The type other event is used for
injection of events that are not delivered through the IDT.
— For exceptions, the deliver-error-code bit (bit 11) determines whether
delivery pushes an error code on the guest stack.
— VM entry injects an event if and only if the valid bit (bit 31) is 1. The valid bit
in this field is cleared on every VM exit (see Section 24.2).
•
VM-entry exception error code (32 bits). This field is used if and only if the
valid bit (bit 31) and the deliver-error-code bit (bit 11) are both set in the
VM-entry interruption-information field.
•
VM-entry instruction length (32 bits). For injection of events whose type is
software interrupt, software exception, or privileged software exception, this
field is used to determine the value of RIP that is pushed on the stack.
See Section 23.5 for details regarding the mechanics of event injection, including the
use of the interruption type and the VM-entry instruction length.
VM exits clear the valid bit (bit 31) in the VM-entry interruption-information field.
21-26 Vol. 3B
VIRTUAL-MACHINE CONTROL STRUCTURES
21.9
VM-EXIT INFORMATION FIELDS
The VMCS contains a section of read-only fields that contain information about the
most recent VM exit. Attempts to write to these fields with VMWRITE fail (see
“VMWRITE—Write Field to Virtual-Machine Control Structure” in Chapter 6 of the
Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 2B).
21.9.1
Basic VM-Exit Information
The following VM-exit information fields provide basic information about a VM exit:
•
Exit reason (32 bits). This field encodes the reason for the VM exit and has the
structure given in Table 21-13.
Table 21-13. Format of Exit Reason
Bit
Position(s)
Contents
15:0
Basic exit reason
27:16
Reserved (cleared to 0)
28
Pending MTF VM exit
29
VM exit from VMX root operation
30
Reserved (cleared to 0)
31
VM-entry failure (0 = true VM exit; 1 = VM-entry failure)
— Bits 15:0 provide basic information about the cause of the VM exit (if bit 31 is
clear) or of the VM-entry failure (if bit 31 is set). Appendix I enumerates the
basic exit reasons.
— Bit 28 is set only by an SMM VM exit (see Section 26.15.2) that took priority
over an MTF VM exit (see Section 22.7.2) that would have occurred had the
SMM VM exit not occurred. See Section 26.15.2.3.
— Bit 29 is set if and only if the processor was in VMX root operation at the time
the VM exit occurred. This can happen only for SMM VM exits. See Section
26.15.2.
— Because some VM-entry failures load processor state from the host-state
area (see Section 23.7), software must be able to distinguish such cases from
true VM exits. Bit 31 is used for that purpose.
•
Exit qualification (64 bits; 32 bits on processors that do not support Intel 64
architecture). This field contains additional information about the cause of
VM exits due to the following: debug exceptions; page-fault exceptions; start-up
IPIs (SIPIs); task switches; INVEPT; INVLPG;INVVPID; LGDT; LIDT; LLDT; LTR;
Vol. 3B 21-27
VIRTUAL-MACHINE CONTROL STRUCTURES
SGDT; SIDT; SLDT; STR; VMCLEAR; VMPTRLD; VMPTRST; VMREAD; VMWRITE;
VMXON; control-register accesses; MOV DR; I/O instructions; and MWAIT. The
format of the field depends on the cause of the VM exit. See Section 24.2.1 for
details.
•
Guest-linear address (64 bits; 32 bits on processors that do not support
Intel 64 architecture). This field is used in the following cases:
— VM exits due to attempts to execute LMSW with a memory operand.
— VM exits due to attempts to execute INS or OUTS.
— VM exits due to system-management interrupts (SMIs) that arrive
immediately after retirement of I/O instructions.
— Certain VM exits due to EPT violations
See Section 24.2.1 and Section 26.15.2.3 for details of when and how this field is
used.
•
Guest-physical address (64 bits). This field is used VM exits due to EPT
violations and EPT misconfigurations. See Section 24.2.1 for details of when and
how this field is used.
21.9.2
Information for VM Exits Due to Vectored Events
Event-specific information is provided for VM exits due to the following vectored
events: exceptions (including those generated by the instructions INT3, INTO,
BOUND, and UD2); external interrupts that occur while the “acknowledge interrupt
on exit” VM-exit control is 1; and non-maskable interrupts (NMIs). This information
is provided in the following fields:
•
VM-exit interruption information (32 bits). This field receives basic
information associated with the event causing the VM exit. Table 21-14 describes
this field.
Table 21-14. Format of the VM-Exit Interruption-Information Field
Bit Position(s) Content
7:0
Vector of interrupt or exception
10:8
Interruption type:
0: External interrupt
1: Not used
2: Non-maskable interrupt (NMI)
3: Hardware exception
4 – 5: Not used
6: Software exception
7: Not used
11
Error code valid (0 = invalid; 1 = valid)
12
NMI unblocking due to IRET
21-28 Vol. 3B
VIRTUAL-MACHINE CONTROL STRUCTURES
Table 21-14. Format of the VM-Exit Interruption-Information Field (Contd.)
Bit Position(s) Content
30:13
Reserved (cleared to 0)
31
Valid
•
VM-exit interruption error code (32 bits). For VM exits caused by hardware
exceptions that would have delivered an error code on the stack, this field
receives that error code.
Section 24.2.2 provides details of how these fields are saved on VM exits.
21.9.3
Information for VM Exits That Occur During Event Delivery
Additional information is provided for VM exits that occur during event delivery in
VMX non-root operation.1 This information is provided in the following fields:
•
IDT-vectoring information (32 bits). This field receives basic information
associated with the event that was being delivered when the VM exit occurred.
Table 21-15 describes this field.
Table 21-15. Format of the IDT-Vectoring Information Field
Bit
Position(s)
Content
7:0
Vector of interrupt or exception
10:8
Interruption type:
0: External interrupt
1: Not used
2: Non-maskable interrupt (NMI)
3: Hardware exception
4: Software interrupt
5: Privileged software exception
6: Software exception
7: Not used
11
Error code valid (0 = invalid; 1 = valid)
12
Undefined
30:13
Reserved (cleared to 0)
31
Valid
1. This includes cases in which the event delivery was caused by event injection as part of
VM entry; see Section 23.5.1.2.
Vol. 3B 21-29
VIRTUAL-MACHINE CONTROL STRUCTURES
•
IDT-vectoring error code (32 bits). For VM exits the occur during delivery of
hardware exceptions that would have delivered an error code on the stack, this
field receives that error code.
See Section 24.2.3 provides details of how these fields are saved on VM exits.
21.9.4
Information for VM Exits Due to Instruction Execution
The following fields are used for VM exits caused by attempts to execute certain
instructions in VMX non-root operation:
•
VM-exit instruction length (32 bits). For VM exits resulting from instruction
execution, this field receives the length in bytes of the instruction whose
execution led to the VM exit.1 See Section 24.2.4 for details of when and how this
field is used.
•
VM-exit instruction information (32 bits). This field is used for VM exits due
to attempts to execute INS, INVEPT, INVVPID, LIDT, LGDT, LLDT, LTR, OUTS,
SIDT, SGDT, SLDT, STR, VMCLEAR, VMPTRLD, VMPTRST, VMREAD, VMWRITE, or
VMXON.2 The format of the field depends on the cause of the VM exit. See
Section 24.2.4 for details.
The following fields (64 bits each; 32 bits on processors that do not support Intel 64
architecture) are used only for VM exits due to SMIs that arrive immediately after
retirement of I/O instructions. They provide information about that I/O instruction:
•
•
•
•
I/O RCX. The value of RCX before the I/O instruction started.
I/O RSI. The value of RSI before the I/O instruction started.
I/O RDI. The value of RDI before the I/O instruction started.
I/O RIP. The value of RIP before the I/O instruction started (the RIP that
addressed the I/O instruction).
21.9.5
VM-Instruction Error Field
The 32-bit VM-instruction error field does not provide information about the most
recent VM exit. In fact, it is not modified on VM exits. Instead, it provides information
about errors encountered by a non-faulting execution of one of the VMX instructions.
1. This field is also used for VM exits that occur during the delivery of a software interrupt or software exception.
2. Whether the processor provides this information on VM exits due to attempts to execute INS or
OUTS can be determined by consulting the VMX capability MSR IA32_VMX_BASIC (see Appendix
G.1).
21-30 Vol. 3B
VIRTUAL-MACHINE CONTROL STRUCTURES
21.10
SOFTWARE USE OF THE VMCS AND RELATED
STRUCTURES
This section details guidelines that software should observe when using a VMCS and
related structures. It also provides descriptions of consequences for failing to follow
guidelines.
21.10.1 Software Use of Virtual-Machine Control Structures
To ensure proper processor behavior, software should observe certain guidelines
when using an active VMCS.
No VMCS should ever be active on more than one logical processor. If a VMCS is to be
“migrated” from one logical processor to another, the first logical processor should
execute VMCLEAR for the VMCS (to make it inactive on that logical processor and to
ensure that all VMCS data are in memory) before the other logical processor
executes VMPTRLD for the VMCS (to make it active on the second logical processor).
A VMCS that is made active on more than one logical processor may become
corrupted (see below).
Software should use the VMREAD and VMWRITE instructions to access the different
fields in the current VMCS (see Section 21.10.2). Software should never access or
modify the VMCS data of an active VMCS using ordinary memory operations, in part
because the format used to store the VMCS data is implementation-specific and not
architecturally defined, and also because a logical processor may maintain some
VMCS data of an active VMCS on the processor and not in the VMCS region. The
following items detail some of the hazards of accessing VMCS data using ordinary
memory operations:
•
Any data read from a VMCS with an ordinary memory read does not reliably
reflect the state of the VMCS. Results may vary from time to time or from logical
processor to logical processor.
•
Writing to a VMCS with an ordinary memory write is not guaranteed to have a
deterministic effect on the VMCS. Doing so may cause the VMCS to become
corrupted (see below).
(Software can avoid these hazards by removing any linear-address mappings to a
VMCS region before executing a VMPTRLD for that region and by not remapping it
until after executing VMCLEAR for that region.)
If a logical processor leaves VMX operation, any VMCSs active on that logical
processor may be corrupted (see below). To prevent such corruption of a VMCS that
may be used either after a return to VMX operation or on another logical processor,
software should VMCLEAR that VMCS before executing the VMXOFF instruction or
removing power from the processor (e.g., as part of a transition to the S3 and S4
power states).
This section has identified operations that may cause a VMCS to become corrupted.
These operations may cause the VMCS’s data to become undefined. Behavior may be
Vol. 3B 21-31
VIRTUAL-MACHINE CONTROL STRUCTURES
unpredictable if that VMCS used subsequently on any logical processor. The following
items detail some hazards of VMCS corruption:
•
VM entries may fail for unexplained reasons or may load undesired processor
state.
•
The processor may not correctly support VMX non-root operation as documented
in Chapter 22 and may generate unexpected VM exits.
•
VM exits may load undesired processor state, save incorrect state into the VMCS,
or cause the logical processor to transition to a shutdown state.
21.10.2 VMREAD, VMWRITE, and Encodings of VMCS Fields
Every field of the VMCS is associated with a 32-bit value that is its encoding. The
encoding is provided in an operand to VMREAD and VMWRITE when software wishes
to read or write that field. These instructions fail if given, in 64-bit mode, an operand
that sets an encoding bit beyond bit 32. See Chapter 5 of the Intel® 64 and IA-32
Architectures Software Developer’s Manual, Volume 2B, for a description of these
instructions.
The structure of the 32-bit encodings of the VMCS components is determined principally by the width of the fields and their function in the VMCS. See Table 21-16.
Table 21-16. Structure of VMCS Component Encoding
Bit Position(s)
Contents
31:15
Reserved (must be 0)
14:13
Width:
0: 16-bit
1: 64-bit
2: 32-bit
3: natural-width
12
Reserved (must be 0)
11:10
Type:
0: control
1: read-only data
2: guest state
3: host state
9:1
Index
0
Access type (0 = full; 1 = high); must be full for 16-bit, 32-bit, and naturalwidth fields
The following items detail the meaning of the bits in each encoding:
21-32 Vol. 3B
VIRTUAL-MACHINE CONTROL STRUCTURES
•
Field width. Bits 14:13 encode the width of the field.
— A value of 0 indicates a 16-bit field.
— A value of 1 indicates a 64-bit field.
— A value of 2 indicates a 32-bit field.
— A value of 3 indicates a natural-width field. Such fields have 64 bits on
processors that support Intel 64 architecture and 32 bits on processors that
do not.
Fields whose encodings use value 1 are specially treated to allow 32-bit software
access to all 64 bits of the field. Such access is allowed by defining, for each such
field, an encoding that allows direct access to the high 32 bits of the field. See
below.
•
Field type. Bits 11:10 encode the type of VMCS field: control, guest-state, hoststate, or read-only data. The last category includes the VM-exit information fields
and the VM-instruction error field.
•
•
Index. Bits 9:1 distinguish components with the same field width and type.
Access type. Bit 0 must be 0 for all fields except for 64-bit fields (those with
field-width 1; see above). A VMREAD or VMWRITE using an encoding with this bit
cleared to 0 accesses the entire field. For a 64-bit field with field-width 1, a
VMREAD or VMWRITE using an encoding with this bit set to 1 accesses only the
high 32 bits of the field.
Appendix H gives the encodings of all fields in the VMCS.
The following describes the operation of VMREAD and VMWRITE based on processor
mode, VMCS-field width, and access type:
•
16-bit fields:
— A VMREAD returns the value of the field in bits 15:0 of the destination
operand; other bits of the destination operand are cleared to 0.
— A VMWRITE writes the value of bits 15:0 of the source operand into the VMCS
field; other bits of the source operand are not used.
•
32-bit fields:
— A VMREAD returns the value of the field in bits 31:0 of the destination
operand; in 64-bit mode, bits 63:32 of the destination operand are cleared to
0.
— A VMWRITE writes the value of bits 31:0 of the source operand into the VMCS
field; in 64-bit mode, bits 63:32 of the source operand are not used.
•
64-bit fields and natural-width fields using the full access type outside IA-32e
mode.
— A VMREAD returns the value of bits 31:0 of the field in its destination
operand; bits 63:32 of the field are ignored.
Vol. 3B 21-33
VIRTUAL-MACHINE CONTROL STRUCTURES
— A VMWRITE writes the value of its source operand to bits 31:0 of the field and
clears bits 63:32 of the field.
•
64-bit fields and natural-width fields using the full access type in 64-bit mode
(only on processors that support Intel 64 architecture).
— A VMREAD returns the value of the field in bits 63:0 of the destination
operand
— A VMWRITE writes the value of bits 63:0 of the source operand into the VMCS
field.
•
64-bit fields using the high access type.
— A VMREAD returns the value of bits 63:32 of the field in bits 31:0 of the
destination operand; in 64-bit mode, bits 63:32 of the destination operand
are cleared to 0.
— A VMWRITE writes the value of bits 31:0 of the source operand to bits 63:32
of the field; in 64-bit mode, bits 63:32 of the source operand are not used.
Software seeking to read a 64-bit field outside IA-32e mode can use VMREAD with
the full access type (reading bits 31:0 of the field) and VMREAD with the high access
type (reading bits 63:32 of the field); the order of the two VMREAD executions is not
important. Software seeking to modify a 64-bit field outside IA-32e mode should first
use VMWRITE with the full access type (establishing bits 31:0 of the field while
clearing bits 63:32) and then use VMWRITE with the high access type (establishing
bits 63:32 of the field).
21.10.3 Initializing a VMCS
Software should initialize fields in a VMCS (using VMWRITE) before using the VMCS
for VM entry. Failure to do so may result in unpredictable behavior; for example, a
VM entry may fail for unexplained reasons, or a successful transition (VM entry or
VM exit) may load processor state with unexpected values.
It is not necessary to initialize fields that the logical processor will not use. (For
example, it is not necessary to unitize the MSR-bitmap address if the “use MSR
bitmaps” VM-execution control is 0.)
A processor maintains some VMCS information that cannot be modified with the
VMWRITE instruction; this includes a VMCS’s launch state (see Section 21.1). Such
information may be stored in the VMCS data portion of a VMCS region. Because the
format of this information is implementation-specific, there is no way for software to
know, when it first allocates a region of memory for use as a VMCS region, how the
processor will determine this information from the contents of the memory region.
In addition to its other functions, the VMCLEAR instruction initializes any implementation-specific information in the VMCS region referenced by its operand. To avoid
the uncertainties of implementation-specific behavior, software should execute
VMCLEAR on a VMCS region before making the corresponding VMCS active with
21-34 Vol. 3B
VIRTUAL-MACHINE CONTROL STRUCTURES
VMPTRLD for the first time. (Figure 21-1 illustrates how execution of VMCLEAR puts
a VMCS into a well-defined state.)
The following software usage is consistent with these limitations:
•
VMCLEAR should be executed for a VMCS before it is used for VM entry for the
first time.
•
VMLAUNCH should be used for the first VM entry using a VMCS after VMCLEAR
has been executed for that VMCS.
•
VMRESUME should be used for any subsequent VM entry using a VMCS (until the
next execution of VMCLEAR for the VMCS).
It is expected that, in general, VMRESUME will have lower latency than VMLAUNCH.
Since “migrating” a VMCS from one logical processor to another requires use of
VMCLEAR (see Section 21.10.1), which sets the launch state of the VMCS to “clear”,
such migration requires the next VM entry to be performed using VMLAUNCH. Software developers can avoid the performance cost of increased VM-entry latency by
avoiding unnecessary migration of a VMCS from one logical processor to another.
21.10.4 Software Access to Related Structures
In addition to data in the VMCS region itself, VMX non-root operation can be
controlled by data structures that are referenced by pointers in a VMCS (for example,
the I/O bitmaps). While the pointers to these data structures are parts of the VMCS,
the data structures themselves are not. They are not accessible using VMREAD and
VMWRITE but by ordinary memory writes.
Software should ensure that each such data structure is modified only when no
logical processor with a current VMCS that references it is in VMX non-root operation.
Doing otherwise may lead to unpredictable behavior (including behaviors identified
in Section 21.10.1).
21.10.5 VMXON Region
Before executing VMXON, software allocates a region of memory (called the VMXON
region)1 that the logical processor uses to support VMX operation. The physical
address of this region (the VMXON pointer) is provided in an operand to VMXON. The
VMXON pointer is subject to the limitations that apply to VMCS pointers:
•
•
The VMXON pointer must be 4-KByte aligned (bits 11:0 must be zero).
The VMXON pointer must not set any bits beyond the processor’s physicaladdress width.2,3
1. The amount of memory required for the VMXON region is the same as that required for a VMCS
region. This size is implementation specific and can be determined by consulting the VMX capability MSR IA32_VMX_BASIC (see Appendix G.1).
Vol. 3B 21-35
VIRTUAL-MACHINE CONTROL STRUCTURES
Before executing VMXON, software should write the VMCS revision identifier (see
Section 21.2) to the VMXON region. It need not initialize the VMXON region in any
other way. Software should use a separate region for each logical processor and
should not access or modify the VMXON region of a logical processor between execution of VMXON and VMXOFF on that logical processor. Doing otherwise may lead to
unpredictable behavior (including behaviors identified in Section 21.10.1).
2. Software can determine a processor’s physical-address width by executing CPUID with
80000008H in EAX. The physical-address width is returned in bits 7:0 of EAX.
3. If IA32_VMX_BASIC[48] is read as 1, the VMXON pointer must not set any bits in the range
63:32; see Appendix G.1.
21-36 Vol. 3B
CHAPTER 22
VMX NON-ROOT OPERATION
In a virtualized environment using VMX, the guest software stack typically runs on a
logical processor in VMX non-root operation. This mode of operation is similar to that
of ordinary processor operation outside of the virtualized environment. This chapter
describes the differences between VMX non-root operation and ordinary processor
operation with special attention to causes of VM exits (which bring a logical processor
from VMX non-root operation to root operation). The differences between VMX nonroot operation and ordinary processor operation are described in the following
sections:
•
•
•
•
•
•
•
Section 22.1, “Instructions That Cause VM Exits”
Section 22.2, “APIC-Access VM Exits”
Section 22.3, “Other Causes of VM Exits”
Section 22.4, “Changes to Instruction Behavior in VMX Non-Root Operation”
Section 22.5, “APIC Accesses That Do Not Cause VM Exits”
Section 22.6, “Other Changes in VMX Non-Root Operation”
Section 22.7, “Features Specific to VMX Non-Root Operation”
Chapter 21, “Virtual-Machine Control Structures,” describes the data control structure that governs VMX operation (root and non-root). Chapter 22, “VMX Non-Root
Operation,” describes the operation of VM entries which allow the processor to transition from VMX root operation to non-root operation.
22.1
INSTRUCTIONS THAT CAUSE VM EXITS
Certain instructions may cause VM exits if executed in VMX non-root operation.
Unless otherwise specified, such VM exits are “fault-like,” meaning that the instruction causing the VM exit does not execute and no processor state is updated by the
instruction. Section 24.1 details architectural state in the context of a VM exit.
Section 22.1.1 defines the prioritization between faults and VM exits for instructions
subject to both. Section 22.1.2 identifies instructions that cause VM exits whenever
they are executed in VMX non-root operation (and thus can never be executed in
VMX non-root operation). Section 22.1.3 identifies instructions that cause VM exits
depending on the settings of certain VM-execution control fields (see Section 21.6).
22.1.1
Relative Priority of Faults and VM Exits
The following principles describe the ordering between existing faults and VM exits:
Vol. 3B 22-1
VMX NON-ROOT OPERATION
•
Certain exceptions have priority over VM exits. These include invalid-opcode
exceptions, faults based on privilege level,1 and general-protection exceptions
that are based on checking I/O permission bits in the task-state segment (TSS).
For example, execution of RDMSR with CPL = 3 generates a general-protection
exception and not a VM exit.2
•
Faults incurred while fetching instruction operands have priority over VM exits
that are conditioned based on the contents of those operands (see LMSW in
Section 22.1.3).
•
VM exits caused by execution of the INS and OUTS instructions (resulting either
because the “unconditional I/O exiting” VM-execution control is 1 or because the
“use I/O bitmaps control is 1) have priority over the following faults:
— A general-protection fault due to the relevant segment (ES for INS; DS for
OUTS unless overridden by an instruction prefix) being unusable
— A general-protection fault due to an offset beyond the limit of the relevant
segment
— An alignment-check exception
•
Fault-like VM exits have priority over exceptions other than those mentioned
above. For example, RDMSR of a non-existent MSR with CPL = 0 generates a
VM exit and not a general-protection exception.
When Section 22.1.2 or Section 22.1.3 (below) identify an instruction execution that
may lead to a VM exit, it is assumed that the instruction does not incur a fault that
takes priority over a VM exit.
22.1.2
Instructions That Cause VM Exits Unconditionally
The following instructions cause VM exits when they are executed in VMX non-root
operation: CPUID, GETSEC,3 INVD, and XSETBV.4 This is also true of instructions
introduced with VMX, which include: INVEPT, INVVPID, VMCALL,5 VMCLEAR,
VMLAUNCH, VMPTRLD, VMPTRST, VMREAD, VMRESUME, VMWRITE, VMXOFF, and
VMXON.
1. These include faults generated by attempts to execute, in virtual-8086 mode, privileged instructions that are not recognized in that mode.
2. MOV DR is an exception to this rule; see Section 22.1.3.
3. An execution of GETSEC in VMX non-root operation causes a VM exit if CR4.SMXE[Bit 14] = 1
regardless of the value of CPL or RAX. An execution of GETSEC causes an invalid-opcode exception (#UD) if CR4.SMXE[Bit 14] = 0.
4. An execution of XSETBV in VMX non-root operation causes a VM exit if CR4.OSXSAVE[Bit 18] =
1 regardless of the value of CPL, RAX, RCX, or RDX. An execution of XSETBV causes an invalidopcode exception (#UD) if CR4.OSXSAVE[Bit 18] = 0.
5. Under the dual-monitor treatment of SMIs and SMM, executions of VMCALL cause SMM VM exits
in VMX root operation outside SMM. See Section 26.15.2.
22-2 Vol. 3B
VMX NON-ROOT OPERATION
22.1.3
Instructions That Cause VM Exits Conditionally
Certain instructions cause VM exits in VMX non-root operation depending on the
setting of the VM-execution controls. The following instructions can cause “fault-like”
VM exits based on the conditions described:
•
CLTS. The CLTS instruction causes a VM exit if the bits in position 3 (corresponding to CR0.TS) are set in both the CR0 guest/host mask and the CR0 read
shadow.
•
HLT. The HLT instruction causes a VM exit if the “HLT exiting” VM-execution
control is 1.
•
IN, INS/INSB/INSW/INSD, OUT, OUTS/OUTSB/OUTSW/OUTSD. The
behavior of each of these instructions is determined by the settings of the
“unconditional I/O exiting” and “use I/O bitmaps” VM-execution controls:
— If both controls are 0, the instruction executes normally.
— If the “unconditional I/O exiting” VM-execution control is 1 and the “use I/O
bitmaps” VM-execution control is 0, the instruction causes a VM exit.
— If the “use I/O bitmaps” VM-execution control is 1, the instruction causes a
VM exit if it attempts to access an I/O port corresponding to a bit set to 1 in
the appropriate I/O bitmap (see Section 21.6.4). If an I/O operation “wraps
around” the 16-bit I/O-port space (accesses ports FFFFH and 0000H), the I/O
instruction causes a VM exit (the “unconditional I/O exiting” VM-execution
control is ignored if the “use I/O bitmaps” VM-execution control is 1).
See Section 22.1.1 for information regarding the priority of VM exits relative to
faults that may be caused by the INS and OUTS instructions.
•
INVLPG. The INVLPG instruction causes a VM exit if the “INVLPG exiting”
VM-execution control is 1.
•
LGDT, LIDT, LLDT, LTR, SGDT, SIDT, SLDT, STR. These instructions cause
VM exits if the “descriptor-table exiting” VM-execution control is 1.1
•
LMSW. In general, the LMSW instruction causes a VM exit if it would write, for
any bit set in the low 4 bits of the CR0 guest/host mask, a value different than the
corresponding bit in the CR0 read shadow. LMSW never clears bit 0 of CR0
(CR0.PE); thus, LMSW causes a VM exit if either of the following are true:
— The bits in position 0 (corresponding to CR0.PE) are set in both the CR0
guest/mask and the source operand, and the bit in position 0 is clear in the
CR0 read shadow.
— For any bit position in the range 3:1, the bit in that position is set in the CR0
guest/mask and the values of the corresponding bits in the source operand
and the CR0 read shadow differ.
1. “Descriptor-table exiting” is a secondary processor-based VM-execution control. If bit 31 of the
primary processor-based VM-execution controls is 0, VMX non-root operation functions as if the
“descriptor-table exiting” VM-execution control were 0. See Section 21.6.2.
Vol. 3B 22-3
VMX NON-ROOT OPERATION
•
MONITOR. The MONITOR instruction causes a VM exit if the “MONITOR exiting”
VM-execution control is 1.
•
MOV from CR3. The MOV from CR3 instruction causes a VM exit if the “CR3store exiting” VM-execution control is 1. The first processors to support the
virtual-machine extensions supported only the 1-setting of this control.
•
MOV from CR8. The MOV from CR8 instruction (which can be executed only in
64-bit mode) causes a VM exit if the “CR8-store exiting” VM-execution control is
1. If this control is 0, the behavior of the MOV from CR8 instruction is modified if
the “use TPR shadow” VM-execution control is 1 (see Section 22.4).
•
MOV to CR0. The MOV to CR0 instruction causes a VM exit unless the value of its
source operand matches, for the position of each bit set in the CR0 guest/host
mask, the corresponding bit in the CR0 read shadow. (If every bit is clear in the
CR0 guest/host mask, MOV to CR0 cannot cause a VM exit.)
•
MOV to CR3. The MOV to CR3 instruction causes a VM exit unless the “CR3-load
exiting” VM-execution control is 0 or the value of its source operand is equal to
one of the CR3-target values specified in the VMCS. If the CR3-target count in n,
only the first n CR3-target values are considered; if the CR3-target count is 0,
MOV to CR3 always causes a VM exit.
The first processors to support the virtual-machine extensions supported only
the 1-setting of the “CR3-load exiting” VM-execution control. These processors
always consult the CR3-target controls to determine whether an execution of
MOV to CR3 causes a VM exit.
•
MOV to CR4. The MOV to CR4 instruction causes a VM exit unless the value of its
source operand matches, for the position of each bit set in the CR4 guest/host
mask, the corresponding bit in the CR4 read shadow.
•
MOV to CR8. The MOV to CR8 instruction (which can be executed only in 64-bit
mode) causes a VM exit if the “CR8-load exiting” VM-execution control is 1. If this
control is 0, the behavior of the MOV to CR8 instruction is modified if the “use TPR
shadow” VM-execution control is 1 (see Section 22.4) and it may cause a traplike VM exit (see below).
•
MOV DR. The MOV DR instruction causes a VM exit if the “MOV-DR exiting”
VM-execution control is 1. Such VM exits represent an exception to the principles
identified in Section 22.1.1 in that they take priority over the following: generalprotection exceptions based on privilege level; and invalid-opcode exceptions
that occur because CR4.DE=1 and the instruction specified access to DR4 or DR5.
•
MWAIT. The MWAIT instruction causes a VM exit if the “MWAIT exiting”
VM-execution control is 1. If this control is 0, the behavior of the MWAIT
instruction may be modified (see Section 22.4).
•
PAUSE.The behavior of each of this instruction depends on CPL and the settings
of the “PAUSE exiting” and “PAUSE-loop exiting” VM-execution controls:
— CPL = 0.
•
22-4 Vol. 3B
If the “PAUSE exiting” and “PAUSE-loop exiting” VM-execution controls
are both 0, the PAUSE instruction executes normally.
VMX NON-ROOT OPERATION
•
If the “PAUSE exiting” VM-execution control is 1, the PAUSE instruction
causes a VM exit (the “PAUSE-loop exiting” VM-execution control is
ignored if CPL = 0 and the “PAUSE exiting” VM-execution control is 1).
•
If the “PAUSE exiting” VM-execution control is 0 and the “PAUSE-loop
exiting” VM-execution control is 1, the following treatment applies.
The logical processor determines the amount of time between this
execution of PAUSE and the previous execution of PAUSE at CPL 0. If this
amount of time exceeds the value of the VM-execution control field
PLE_Gap, the processor considers this execution to be the first execution
of PAUSE in a loop. (It also does so for the first execution of PAUSE at CPL
0 after VM entry.)
Otherwise, the logical processor determines the amount of time since the
most recent execution of PAUSE that was considered to be the first in a
loop. If this amount of time exceeds the value of the VM-execution control
field PLE_Window, a VM exit occurs.
For purposes of these computations, time is measured based on a counter
that runs at the same rate as the timestamp counter (TSC).
— CPL > 0.
•
If the “PAUSE exiting” VM-execution control is 0, the PAUSE instruction
executes normally.
•
If the “PAUSE exiting” VM-execution control is 1, the PAUSE instruction
causes a VM exit.
The “PAUSE-loop exiting” VM-execution control is ignored if CPL > 0.
•
RDMSR. The RDMSR instruction causes a VM exit if any of the following are true:
— The “use MSR bitmaps” VM-execution control is 0.
— The value of ECX is not in the range 00000000H – 00001FFFH or
C0000000H – C0001FFFH.
— The value of ECX is in the range 00000000H – 00001FFFH and bit n in read
bitmap for low MSRs is 1, where n is the value of ECX.
— The value of ECX is in the range C0000000H – C0001FFFH and bit n in read
bitmap for high MSRs is 1, where n is the value of ECX & 00001FFFH.
See Section 21.6.9 for details regarding how these bitmaps are identified.
•
RDPMC. The RDPMC instruction causes a VM exit if the “RDPMC exiting”
VM-execution control is 1.
•
RDTSC. The RDTSC instruction causes a VM exit if the “RDTSC exiting”
VM-execution control is 1.
•
RDTSCP. The RDTSCP instruction causes a VM exit if the “RDTSC exiting” and
“enable RDTSCP” VM-execution controls are both 1.
•
RSM. The RSM instruction causes a VM exit if executed in system-management
mode (SMM).1
Vol. 3B 22-5
VMX NON-ROOT OPERATION
•
WBINVD. The WBINVD instruction causes a VM exit if the “WBINVD exiting”
VM-execution control is 1.1
•
WRMSR. The WRMSR instruction causes a VM exit if any of the following are
true:
— The “use MSR bitmaps” VM-execution control is 0.
— The value of ECX is not in the range 00000000H – 00001FFFH or
C0000000H – C0001FFFH.
— The value of ECX is in the range 00000000H – 00001FFFH and bit n in write
bitmap for low MSRs is 1, where n is the value of ECX.
— The value of ECX is in the range C0000000H – C0001FFFH and bit n in write
bitmap for high MSRs is 1, where n is the value of ECX & 00001FFFH.
See Section 21.6.9 for details regarding how these bitmaps are identified.
If an execution of WRMSR does not cause a VM exit as specified above and
ECX = 808H (indicating the TPR MSR), instruction behavior is modified if the
“virtualize x2APIC mode” VM-execution control is 1 (see Section 22.4) and it
may cause a trap-like VM exit (see below).2
The MOV to CR8 and WRMSR instructions may cause “trap-like” VM exits. In such a
case, the instruction completes before the VM exit occurs and that processor state is
updated by the instruction (for example, the value of CS:RIP saved in the guest-state
area of the VMCS references the next instruction).
Specifically, a trap-like VM exit occurs following either instruction if the execution
reduces the value of the TPR shadow below that of the TPR threshold VM-execution
control field (see Section 21.6.8 and Section 22.4) and the following hold:
•
For MOV to CR8:
— The “CR8-load exiting” VM-execution control is 0.
— The “use TPR shadow” VM-execution control is 1.
•
For WRMSR:
— The “use MSR bitmaps” VM-execution control is 1, the value of ECX is 808H,
and bit 808H in write bitmap for low MSRs is 0 (see above).
— The “virtualize x2APIC mode” VM-execution control is 1.
1. Execution of the RSM instruction outside SMM causes an invalid-opcode exception regardless of
whether the processor is in VMX operation. It also does so in VMX root operation in SMM; see
Section 26.15.3.
1. “WBINVD exiting” is a secondary processor-based VM-execution control. If bit 31 of the primary
processor-based VM-execution controls is 0, VMX non-root operation functions as if the
“WBINVD exiting” VM-execution control were 0. See Section 21.6.2.
2. “Virtualize x2APIC mode” is a secondary processor-based VM-execution control. If bit 31 of the
primary processor-based VM-execution controls is 0, VMX non-root operation functions as if the
“virtualize x2APIC mode” VM-execution control were 0. See Section 21.6.2.
22-6 Vol. 3B
VMX NON-ROOT OPERATION
22.2
APIC-ACCESS VM EXITS
If the “virtualize APIC accesses” VM-execution control is 1, an attempt to access
memory using a physical address on the APIC-access page (see Section 21.6.8)
causes a VM exit.1,2 Such a VM exit is called an APIC-access VM exit.
Whether an operation that attempts to access memory with a physical address on the
APIC-access page causes an APIC-access VM exit may be qualified based on the type
of access. Section 22.2.1 describes the treatment of linear accesses, while Section
22.2.3 describes that of physical accesses. Section 22.2.4 discusses accesses to the
TPR field on the APIC-access page (called VTPR accesses), which do not, if the “use
TPR shadow” VM-execution control is 1, cause APIC-access VM exits.
22.2.1
Linear Accesses to the APIC-Access Page
An access to the APIC-access page is called a linear access if (1) it results from a
memory access using a linear address; and (2) the access’s physical address is the
translation of that linear address. Section 22.2.1.1 specifies which linear accesses to
the APIC-access page cause APIC-access VM exits.
In general, the treatment of APIC-access VM exits caused by linear accesses is
similar to that of page faults and EPT violations. Based upon this treatment, Section
22.2.1.2 specifies the priority of such VM exits with respect to other events, while
Section 22.2.1.3 discusses instructions that may cause page faults without accessing
memory and the treatment when they access the APIC-access page.
22.2.1.1
Linear Accesses That Cause APIC-Access VM Exits
Whether a linear access to the APIC-access page causes an APIC-access VM exit
depends in part of the nature of the translation used by the linear address:
•
If the linear access uses a translation with a 4-KByte page, it causes an APICaccess VM exit.
•
If the linear access uses a translation with a large page (2-MByte, 4-MByte, or
1-GByte), the access may or may not cause an APIC-access VM exit. Section
22.5.1 describes the treatment of such accesses that do not cause an APICaccess VM exits.
1. “Virtualize APIC accesses” is a secondary processor-based VM-execution control. If bit 31 of the
primary processor-based VM-execution controls is 0, VMX non-root operation functions as if the
“virtualize APIC accesses” VM-execution control were 0. See Section 21.6.2.
2. Even when addresses are translated using EPT (see Section 25.2), the determination of whether
an APIC-access VM exit occurs depends on an access’s physical address, not its guest-physical
address.
Vol. 3B 22-7
VMX NON-ROOT OPERATION
If CR0.PG = 1 and EPT is in use (the “enable EPT” VM-execution control is 1), a
linear access uses a translation with a large page only if a large page is specified
by both the guest paging structures and the EPT paging structures.1
It is recommended that software configure the paging structures so that any translation to the APIC-access page uses a 4-KByte page.
A linear access to the APIC-access page might not cause an APIC-access VM exit if
the “enable EPT” VM-execution control is 1 and software has not properly invalidate
information cached from the EPT paging structures:
•
At time t1, EPT was in use, the EPTP value was X, and some guest-physical
address Y translated to an address that was not on the APIC-access page at that
time. (This might be because the “virtualize APIC accesses” VM-execution control
was 0.)
•
At later time t2, EPT is in use, the EPTP value is X, and a memory access uses a
linear address that translates to Y, which now translates to an address on the
APIC-access page. (This implies that the “virtualize APIC accesses” VM-execution
control is 1 at this time.)
•
Software did not execute the INVEPT instruction between times t1 and t2, either
with the all-context INVEPT type or with the single-context INVEPT type and X as
the INVEPT descriptor.
In this case, the linear access at time t2 might or might not cause an APIC-access
VM exit. If it does not, the access operates on memory on the APIC-access page.
Software can avoid this situation through appropriate use of the INVEPT instruction;
see Section 25.3.3.4.
A linear access to the APIC-access page might not cause an APIC-access VM exit if
the “enable VPID” VM-execution control is 1 and software has not properly invalidated the TLBs and paging-structure caches:
•
At time t1, the processor was in VMX non-root operation with non-zero VPID X,
and some linear address Y translated to an address that was not on the APICaccess page at that time. (This might be because the “virtualize APIC accesses”
VM-execution control was 0.)
•
At later time t2, the processor was again in VMX non-root operation with VPID X,
and a memory access uses linear address, which now translates to an address on
the APIC-access page. (This implies that the “virtualize APIC accesses” VMexecution control is 1 at this time.)
•
Software did not execute the INVVPID instruction in any of the following ways
between times t1 and t2:
1. If the capability MSR IA32_VMX_CR0_FIXED0 reports that CR0.PG must be 1 in VMX operation,
CR0.PG must be 1 unless the “unrestricted guest” VM-execution control and bit 31 of the primary
processor-based VM-execution controls are both 1. “Enable EPT” is a secondary processor-based
VM-execution control. If bit 31 of the primary processor-based VM-execution controls is 0, VMX
non-root operation functions as if the “enable EPT” VM-execution control were 0. See Section
21.6.2.
22-8 Vol. 3B
VMX NON-ROOT OPERATION
— With the individual-address INVVPID type and an INVVPID descriptor
specifying VPID X and linear address Y.
— With the single-context INVVPID type and an INVVPID descriptor specifying
VPID X.
— With the all-context INVEPT type.
— With the single-context-retaining-globals INVVPID type and an INVVPID
descriptor specifying VPID X (assuming that, at time t1, the translation for Y
was global; see Section 4.10, “Caching Translation Information” in Intel® 64
and IA-32 Architectures Software Developer’s Manual, Volume 3A for details
regarding global translations).
In this case, the linear access at time t2 might or might not cause an APIC-access
VM exit. If it does not, the access operates on memory on the APIC-access page.
Software can avoid this situation through appropriate use of the INVVPID instruction;
see Section 25.3.3.3.
22.2.1.2
Priority of APIC-Access VM Exits Caused by Linear Accesses
The following items specify the priority relative to other events of APIC-access
VM exits caused by linear accesses.
•
The priority of an APIC-access VM exit on a linear access to memory is below that
of any page fault or EPT violation that that access may incur. That is, a linear
access does not cause an APIC-access VM exit if it would cause a page fault or an
EPT violation.
•
A linear access does not cause an APIC-access VM exit until after the accessed
bits are set in the paging structures.
•
A linear write access will not cause an APIC-access VM exit until after the dirty bit
is set in the appropriate paging structure.
•
With respect to all other events, any APIC-access VM exit due to a linear access
has the same priority as any page fault or EPT violation that the linear access
could cause. (This item applies to other events that the linear access may
generate as well as events that may be generated by other accesses by the same
instruction or operation.)
These principles imply among other things, that an APIC-access VM exit may occur
during the execution of a repeated string instruction (including INS and OUTS).
Suppose, for example, that the first n iterations (n may be 0) of such an instruction
do not access the APIC-access page and that the next iteration does access that
page. As a result, the first n iterations may complete and be followed by an APICaccess VM exit. The instruction pointer saved in the VMCS references the repeated
string instruction and the values of the general-purpose registers reflect the completion of n iterations.
Vol. 3B 22-9
VMX NON-ROOT OPERATION
22.2.1.3
Instructions That May Cause Page Faults or EPT Violations
Without Accessing Memory
APIC-access VM exits may occur as a result of executing an instruction that can
cause a page fault or an EPT violation even if that instruction would not access the
APIC-access page. The following are some examples:
•
The CLFLUSH instruction is considered to read from the linear address in its
source operand. If that address translates to one on the APIC-access page, the
instruction causes an APIC-access VM exit.
•
The ENTER instruction causes a page fault if the byte referenced by the final
value of the stack pointer is not writable (even though ENTER does not write to
that byte if its size operand is non-zero). If that byte is writable but is on the
APIC-access page, ENTER causes an APIC-access VM exit.1
•
An execution of the MASKMOVQ or MASKMOVDQU instructions with a zero mask
may or may not cause a page fault or an EPT violation if the destination page is
unwritable (the behavior is implementation-specific). An execution with a zero
mask causes an APIC-access VM exit only on processors for which it could cause
a page fault or an EPT violation.
•
The MONITOR instruction is considered to read from the effective address in RAX.
If the linear address corresponding to that address translates to one on the APICaccess page, the instruction causes an APIC-access VM exit.2
•
An execution of the PREFETCH instruction that would result in an access to the
APIC-access page does not cause an APIC-access VM exit.
22.2.2
Guest-Physical Accesses to the APIC-Access Page
An access to the APIC-access page is called a guest-physical access if
(1) CR0.PG = 1;3 (2) the “enable EPT” VM-execution control is 1;4 (3) the access’s
physical address is the result of an EPT translation; and (4) either (a) the access was
1. The ENTER instruction may also cause page faults due to the memory accesses that it actually
does perform. With regard to APIC-access VM exits, these are treated just as accesses by any
other instruction.
2. This chapter uses the notation RAX, RIP, RSP, RFLAGS, etc. for processor registers because most
processors that support VMX operation also support Intel 64 architecture. For IA-32 processors,
this notation refers to the 32-bit forms of those registers (EAX, EIP, ESP, EFLAGS, etc.). In a few
places, notation such as EAX is used to refer specifically to lower 32 bits of the indicated register.
3. If the capability MSR IA32_VMX_CR0_FIXED0 reports that CR0.PG must be 1 in VMX operation,
CR0.PG must be 1 unless the “unrestricted guest” VM-execution control and bit 31 of the primary
processor-based VM-execution controls are both 1.
4. “Enable EPT” is a secondary processor-based VM-execution control. If bit 31 of the primary processor-based VM-execution controls is 0, VMX non-root operation functions as if the “enable
EPT” VM-execution control were 0. See Section 21.6.2.
22-10 Vol. 3B
VMX NON-ROOT OPERATION
not generated by a linear address; or (b) the access’s guest-physical address is not
the translation of the access’s linear address. Guest-physical accesses include the
following when guest-physical addresses are being translated using EPT:
•
Reads from the guest paging structures when translating a linear address (such
an access uses a guest-physical address that is not the translation of that linear
address).
•
Loads of the page-directory-pointer-table entries by MOV to CR when the logical
processor is using (or that causes the logical processor to use) PAE paging.1
•
Updates to the accessed and dirty bits in the guest paging structures when using
a linear address (such an access uses a guest-physical address that is not the
translation of that linear address).
Section 22.2.2.1 specifies when guest-physical accesses to the APIC-access page
might not cause APIC-access VM exits. In general, the treatment of APIC-access
VM exits caused by guest-physical accesses is similar to that of EPT violations. Based
upon this treatment, Section 22.2.2.2 specifies the priority of such VM exits with
respect to other events.
22.2.2.1
Guest-Physical Accesses That Might Not Cause APIC-Access
VM Exits
Whether a guest-physical access to the APIC-access page causes an APIC-access
VM exit depends on the nature of the EPT translation used by the guest-physical
address and on how software is managing information cached from the EPT paging
structures. The following items detail cases in which a guest-physical access to the
APIC-access page might not cause an APIC-access VM exit:
•
If the access uses a guest-physical address whose translation to the APIC-access
page uses an EPT PDPTE that maps a 1-GByte page (because bit 7 of the EPT
PDPTE is 1).
•
If the access uses a guest-physical address whose translation to the APIC-access
page uses an EPT PDE that maps a 2-MByte page (because bit 7 of the EPT PDE
is 1).
•
Software has not properly invalidated information cached from the EPT paging
structures:
— At time t1, EPT was in use, the EPTP value was X, and some guest-physical
address Y translated to an address that was not on the APIC-access page at
that time. (This might be because the “virtualize APIC accesses” VMexecution control was 0.)
— At later time t2, the EPTP value is X and a memory access uses guest-physical
address Y, which now translates to an address on the APIC-access page. (This
1. A logical processor uses PAE paging if CR0.PG = 1, CR4.PAE = 1 and IA32_EFER.LMA = 0. See
Section 4.4 in the Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 3A.
Vol. 3B 22-11
VMX NON-ROOT OPERATION
implies that the “virtualize APIC accesses” VM-execution control is 1 at this
time.)
— Software did not execute the INVEPT instruction, either with the all-context
INVEPT type or with the single-context INVEPT type and X as the INVEPT
descriptor, between times t1 and t2.
In any of the above cases, the guest-physical access at time t2 might or might not an
APIC-access VM exit. If it does not, the access operates on memory on the APICaccess page.
Software can avoid this situation through appropriate use of the INVEPT instruction;
see Section 25.3.3.4.
22.2.2.2
Priority of APIC-Access VM Exits Caused by Guest-Physical
Accesses
The following items specify the priority relative to other events of APIC-access
VM exits caused by guest-physical accesses.
•
The priority of an APIC-access VM exit caused by a guest-physical access to
memory is below that of any EPT violation that that access may incur. That is, a
guest-physical access does not cause an APIC-access VM exit if it would cause an
EPT violation.
•
With respect to all other events, any APIC-access VM exit caused by a guestphysical access has the same priority as any EPT violation that the guest-physical
access could cause.
22.2.3
Physical Accesses to the APIC-Access Page
An access to the APIC-access page is called a physical access if (1) either (a) the
“enable EPT” VM-execution control is 0;1 or (b) the access’s physical address is not
the result of a translation through the EPT paging structures; and (2) either (a) the
access is not generated by a linear address; or (b) the access’s physical address is
not the translation of its linear address.
Physical accesses include the following:
•
If the “enable EPT” VM-execution control is 0:
— Reads from the paging structures when translating a linear address.
— Loads of the page-directory-pointer-table entries by MOV to CR when the
logical processor is using (or that causes the logical processor to use) PAE
paging.2
1. “Enable EPT” is a secondary processor-based VM-execution control. If bit 31 of the primary processor-based VM-execution controls is 0, VMX non-root operation functions as if the “enable
EPT” VM-execution control were 0. See Section 21.6.2.
22-12 Vol. 3B
VMX NON-ROOT OPERATION
— Updates to the accessed and dirty bits in the paging structures.
•
If the “enable EPT” VM-execution control is 1, accesses to the EPT paging
structures.
•
Any of the following accesses made by the processor to support VMX non-root
operation:
— Accesses to the VMCS region.
— Accesses to data structures referenced (directly or indirectly) by physical
addresses in VM-execution control fields in the VMCS. These include the I/O
bitmaps, the MSR bitmaps, and the virtual-APIC page.
•
Accesses that effect transitions into and out of SMM.1 These include the
following:
— Accesses to SMRAM during SMI delivery and during execution of RSM.
— Accesses during SMM VM exits (including accesses to MSEG) and during
VM entries that return from SMM.
A physical access to the APIC-access page may or may not cause an APIC-access
VM exit. (A physical write to the APIC-access page may write to memory as specified
in Section 22.5.2 before causing the APIC-access VM exit.) The priority of an APICaccess VM exit caused by physical access is not defined relative to other events that
the access may cause. Section 22.5.2 describes the treatment of physical accesses to
the APIC-access page that do not cause APIC-access VM exits.
It is recommended that software not set the APIC-access address to any of those
used by physical memory accesses (identified above). For example, it should not set
the APIC-access address to the physical address of any of the active paging structures if the “enable EPT” VM-execution control is 0.
22.2.4
VTPR Accesses
A memory access is a VTPR access if all of the following hold: (1) the “use TPR
shadow” VM-execution control is 1; (2) the access is not for an instruction fetch;
(3) the access is at most 32 bits in width; and (4) the access is to offset 80H on the
APIC-access page.
A memory access is not a VTPR access (even if it accesses only bytes in the range
80H–83H on the APIC-access page) if any of the following hold: (1) the “use TPR
shadow” VM-execution control is 0; (2) the access is for an instruction fetch; (3) the
access is more than 32 bits in width; or (4) the access is to some offset is on the
APIC-access page other than 80H. For example, a 16-bit access to offset 81H on the
2. A logical processor uses PAE paging if CR0.PG = 1, CR4.PAE = 1 and IA32_EFER.LMA = 0. See
Section 4.4 in the Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 3A.
1. Technically, these accesses do not occur in VMX non-root operation. They are included here for
clarity.
Vol. 3B 22-13
VMX NON-ROOT OPERATION
APIC-access page is not a VTPR access, even if the “use TPR shadow” VM-execution
control is 1.
In general, VTPR accesses do not cause APIC-access VM exits. Instead, they are
treated as described in Section 22.5.3. Physical VTPR accesses (see Section 22.2.3)
may or may not cause APIC-access VM exits; see Section 22.5.2.
22.3
OTHER CAUSES OF VM EXITS
In addition to VM exits caused by instruction execution, the following events can
cause VM exits:
•
Exceptions. Exceptions (faults, traps, and aborts) cause VM exits based on the
exception bitmap (see Section 21.6.3). If an exception occurs, its vector (in the
range 0–31) is used to select a bit in the exception bitmap. If the bit is 1, a
VM exit occurs; if the bit is 0, the exception is delivered normally through the
guest IDT. This use of the exception bitmap applies also to exceptions generated
by the instructions INT3, INTO, BOUND, and UD2.
Page faults (exceptions with vector 14) are specially treated. When a page fault
occurs, a logical processor consults (1) bit 14 of the exception bitmap; (2) the
error code produced with the page fault [PFEC]; (3) the page-fault error-code
mask field [PFEC_MASK]; and (4) the page-fault error-code match field
[PFEC_MATCH]. It checks if PFEC & PFEC_MASK = PFEC_MATCH. If there is
equality, the specification of bit 14 in the exception bitmap is followed (for
example, a VM exit occurs if that bit is set). If there is inequality, the meaning of
that bit is reversed (for example, a VM exit occurs if that bit is clear).
Thus, if software desires VM exits on all page faults, it can set bit 14 in the
exception bitmap to 1 and set the page-fault error-code mask and match fields
each to 00000000H. If software desires VM exits on no page faults, it can set bit
14 in the exception bitmap to 1, the page-fault error-code mask field to
00000000H, and the page-fault error-code match field to FFFFFFFFH.
•
Triple fault. A VM exit occurs if the logical processor encounters an exception
while attempting to call the double-fault handler and that exception itself does
not cause a VM exit due to the exception bitmap. This applies to the case in which
the double-fault exception was generated within VMX non-root operation, the
case in which the double-fault exception was generated during event injection by
VM entry, and to the case in which VM entry is injecting a double-fault exception.
•
External interrupts. An external interrupt causes a VM exit if the “externalinterrupt exiting” VM-execution control is 1. Otherwise, the interrupt is delivered
normally through the IDT. (If a logical processor is in the shutdown state or the
wait-for-SIPI state, external interrupts are blocked. The interrupt is not delivered
through the IDT and no VM exit occurs.)
•
Non-maskable interrupts (NMIs). An NMI causes a VM exit if the “NMI
exiting” VM-execution control is 1. Otherwise, it is delivered using descriptor 2 of
22-14 Vol. 3B
VMX NON-ROOT OPERATION
the IDT. (If a logical processor is in the wait-for-SIPI state, NMIs are blocked. The
NMI is not delivered through the IDT and no VM exit occurs.)
•
INIT signals. INIT signals cause VM exits. A logical processor performs none of
the operations normally associated with these events. Such exits do not modify
register state or clear pending events as they would outside of VMX operation. (If
a logical processor is in the wait-for-SIPI state, INIT signals are blocked. They do
not cause VM exits in this case.)
•
Start-up IPIs (SIPIs). SIPIs cause VM exits. If a logical processor is not in
the wait-for-SIPI activity state when a SIPI arrives, no VM exit occurs and the
SIPI is discarded. VM exits due to SIPIs do not perform any of the normal
operations associated with those events: they do not modify register state as
they would outside of VMX operation. (If a logical processor is not in the wait-forSIPI state, SIPIs are blocked. They do not cause VM exits in this case.)
•
Task switches. Task switches are not allowed in VMX non-root operation. Any
attempt to effect a task switch in VMX non-root operation causes a VM exit. See
Section 22.6.2.
•
System-management interrupts (SMIs). If the logical processor is using the
dual-monitor treatment of SMIs and system-management mode (SMM), SMIs
cause SMM VM exits. See Section 26.15.2.1
•
VMX-preemption timer. A VM exit occurs when the timer counts down to zero.
See Section 22.7.1 for details of operation of the VMX-preemption timer. As noted
in that section, the timer does not cause VM exits if the logical processor is
outside the C-states C0, C1, and C2.
Debug-trap exceptions and higher priority events take priority over VM exits
caused by the VMX-preemption timer. VM exits caused by the VMX-preemption
timer take priority over VM exits caused by the “NMI-window exiting”
VM-execution control and lower priority events.
These VM exits wake a logical processor from the same inactive states as would
a non-maskable interrupt. Specifically, they wake a logical processor from the
shutdown state and from the states entered using the HLT and MWAIT instructions. These VM exits do not occur if the logical processor is in the wait-for-SIPI
state.
In addition, there are controls that cause VM exits based on the readiness of guest
software to receive interrupts:
•
If the “interrupt-window exiting” VM-execution control is 1, a VM exit occurs
before execution of any instruction if RFLAGS.IF = 1 and there is no blocking of
events by STI or by MOV SS (see Table 21-3). Such a VM exit occurs immediately
after VM entry if the above conditions are true (see Section 23.6.5).
1. Under the dual-monitor treatment of SMIs and SMM, SMIs also cause SMM VM exits if they occur
in VMX root operation outside SMM. If the processor is using the default treatment of SMIs and
SMM, SMIs are delivered as described in Section 26.14.1.
Vol. 3B 22-15
VMX NON-ROOT OPERATION
Non-maskable interrupts (NMIs) and higher priority events take priority over
VM exits caused by this control. VM exits caused by this control take priority over
external interrupts and lower priority events.
These VM exits wake a logical processor from the same inactive states as would
an external interrupt. Specifically, they wake a logical processor from the states
entered using the HLT and MWAIT instructions. These VM exits do not occur if the
logical processor is in the shutdown state or the wait-for-SIPI state.
•
If the “NMI-window exiting” VM-execution control is 1, a VM exit occurs before
execution of any instruction if there is no virtual-NMI blocking and there is no
blocking of events by MOV SS (see Table 21-3). (A logical processor may also
prevent such a VM exit if there is blocking of events by STI.) Such a VM exit
occurs immediately after VM entry if the above conditions are true (see Section
23.6.6).
VM exits caused by the VMX-preemption timer and higher priority events take
priority over VM exits caused by this control. VM exits caused by this control take
priority over non-maskable interrupts (NMIs) and lower priority events.
These VM exits wake a logical processor from the same inactive states as would
an NMI. Specifically, they wake a logical processor from the shutdown state and
from the states entered using the HLT and MWAIT instructions. These VM exits do
not occur if the logical processor is in the wait-for-SIPI state.
22.4
CHANGES TO INSTRUCTION BEHAVIOR IN VMX NONROOT OPERATION
The behavior of some instructions is changed in VMX non-root operation. Some of
these changes are determined by the settings of certain VM-execution control fields.
The following items detail such changes:
•
CLTS. Behavior of the CLTS instruction is determined by the bits in position 3
(corresponding to CR0.TS) in the CR0 guest/host mask and the CR0 read
shadow:
— If bit 3 in the CR0 guest/host mask is 0, CLTS clears CR0.TS normally (the
value of bit 3 in the CR0 read shadow is irrelevant in this case), unless CR0.TS
is fixed to 1 in VMX operation (see Section 20.8), in which case CLTS causes
a general-protection exception.
— If bit 3 in the CR0 guest/host mask is 1 and bit 3 in the CR0 read shadow is 0,
CLTS completes but does not change the contents of CR0.TS.
— If the bits in position 3 in the CR0 guest/host mask and the CR0 read shadow
are both 1, CLTS causes a VM exit (see Section 22.1.3).
•
IRET. Behavior of IRET with regard to NMI blocking (see Table 21-3) is
determined by the settings of the “NMI exiting” and “virtual NMIs” VM-execution
controls:
22-16 Vol. 3B
VMX NON-ROOT OPERATION
— If the “NMI exiting” VM-execution control is 0, IRET operates normally and
unblocks NMIs. (If the “NMI exiting” VM-execution control is 0, the “virtual
NMIs” control must be 0; see Section 23.2.1.1.)
— If the “NMI exiting” VM-execution control is 1, IRET does not affect blocking
of NMIs. If, in addition, the “virtual NMIs” VM-execution control is 1, the
logical processor tracks virtual-NMI blocking. In this case, IRET removes any
virtual-NMI blocking.
The unblocking of NMIs or virtual NMIs specified above occurs even if IRET
causes a fault.
•
LMSW. Outside of VMX non-root operation, LMSW loads its source operand into
CR0[3:0], but it does not clear CR0.PE if that bit is set. In VMX non-root
operation, an execution of LMSW that does not cause a VM exit (see Section
22.1.3) leaves unmodified any bit in CR0[3:0] corresponding to a bit set in the
CR0 guest/host mask. An attempt to set any other bit in CR0[3:0] to a value not
supported in VMX operation (see Section 20.8) causes a general-protection
exception. Attempts to clear CR0.PE are ignored without fault.
•
MOV from CR0. The behavior of MOV from CR0 is determined by the CR0
guest/host mask and the CR0 read shadow. For each position corresponding to a
bit clear in the CR0 guest/host mask, the destination operand is loaded with the
value of the corresponding bit in CR0. For each position corresponding to a bit set
in the CR0 guest/host mask, the destination operand is loaded with the value of
the corresponding bit in the CR0 read shadow. Thus, if every bit is cleared in the
CR0 guest/host mask, MOV from CR0 reads normally from CR0; if every bit is set
in the CR0 guest/host mask, MOV from CR0 returns the value of the CR0 read
shadow.
Depending on the contents of the CR0 guest/host mask and the CR0 read
shadow, bits may be set in the destination that would never be set when reading
directly from CR0.
•
MOV from CR3. If the “enable EPT” VM-execution control is 1 and an execution
of MOV from CR3 does not cause a VM exit (see Section 22.1.3), the value loaded
from CR3 is a guest-physical address; see Section 25.2.1.
•
MOV from CR4. The behavior of MOV from CR4 is determined by the CR4
guest/host mask and the CR4 read shadow. For each position corresponding to a
bit clear in the CR4 guest/host mask, the destination operand is loaded with the
value of the corresponding bit in CR4. For each position corresponding to a bit set
in the CR4 guest/host mask, the destination operand is loaded with the value of
the corresponding bit in the CR4 read shadow. Thus, if every bit is cleared in the
CR4 guest/host mask, MOV from CR4 reads normally from CR4; if every bit is set
in the CR4 guest/host mask, MOV from CR4 returns the value of the CR4 read
shadow.
Depending on the contents of the CR4 guest/host mask and the CR4 read
shadow, bits may be set in the destination that would never be set when reading
directly from CR4.
Vol. 3B 22-17
VMX NON-ROOT OPERATION
•
MOV from CR8. Behavior of the MOV from CR8 instruction (which can be
executed only in 64-bit mode) is determined by the settings of the “CR8-store
exiting” and “use TPR shadow” VM-execution controls:
— If both controls are 0, MOV from CR8 operates normally.
— If the “CR8-store exiting” VM-execution control is 0 and the “use TPR
shadow” VM-execution control is 1, MOV from CR8 reads from the TPR
shadow. Specifically, it loads bits 3:0 of its destination operand with the value
of bits 7:4 of byte 80H of the virtual-APIC page (see Section 21.6.8). Bits
63:4 of the destination operand are cleared.
— If the “CR8-store exiting” VM-execution control is 1, MOV from CR8 causes a
VM exit (see Section 22.1.3); the “use TPR shadow” VM-execution control is
ignored in this case.
•
MOV to CR0. An execution of MOV to CR0 that does not cause a VM exit (see
Section 22.1.3) leaves unmodified any bit in CR0 corresponding to a bit set in the
CR0 guest/host mask. Treatment of attempts to modify other bits in CR0 depends
on the setting of the “unrestricted guest” VM-execution control:1
— If the control is 0, MOV to CR0 causes a general-protection exception if it
attempts to set any bit in CR0 to a value not supported in VMX operation (see
Section 20.8).
— If the control is 1, MOV to CR0 causes a general-protection exception if it
attempts to set any bit in CR0 other than bit 0 (PE) or bit 31 (PG) to a value
not supported in VMX operation. It remains the case, however, that MOV to
CR0 causes a general-protection exception if it would result in CR0.PE = 0
and CR0.PG = 1 or if it would result in CR0.PG = 1, CR4.PAE = 0, and
IA32_EFER.LME = 1.
•
MOV to CR3. If the “enable EPT” VM-execution control is 1 and an execution of
MOV to CR3 does not cause a VM exit (see Section 22.1.3), the value loaded into
CR3 is treated as a guest-physical address; see Section 25.2.1.
— If PAE paging is not being used, the instruction does not use the guestphysical address to access memory and it does not cause it to be translated
through EPT.2
— If PAE paging is being used, the instruction translates the guest-physical
address through EPT and uses the result to load the four (4) page-directorypointer-table entries (PDPTEs). The instruction does not use the guestphysical addresses the PDPTEs to access memory and it does not cause them
to be translated through EPT.
1. “Unrestricted guest” is a secondary processor-based VM-execution control. If bit 31 of the primary processor-based VM-execution controls is 0, VMX non-root operation functions as if the
“unrestricted guest” VM-execution control were 0. See Section 21.6.2.
2. A logical processor uses PAE paging if CR0.PG = 1, CR4.PAE = 1 and IA32_EFER.LMA = 0. See
Section 4.4 in the Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 3A.
22-18 Vol. 3B
VMX NON-ROOT OPERATION
•
MOV to CR4. An execution of MOV to CR4 that does not cause a VM exit (see
Section 22.1.3) leaves unmodified any bit in CR4 corresponding to a bit set in the
CR4 guest/host mask. Such an execution causes a general-protection exception
if it attempts to set any bit in CR4 (not corresponding to a bit set in the CR4
guest/host mask) to a value not supported in VMX operation (see Section 20.8).
•
MOV to CR8. Behavior of the MOV to CR8 instruction (which can be executed
only in 64-bit mode) is determined by the settings of the “CR8-load exiting” and
“use TPR shadow” VM-execution controls:
— If both controls are 0, MOV to CR8 operates normally.
— If the “CR8-load exiting” VM-execution control is 0 and the “use TPR shadow”
VM-execution control is 1, MOV to CR8 writes to the TPR shadow. Specifically,
it stores bits 3:0 of its source operand into bits 7:4 of byte 80H of the virtualAPIC page (see Section 21.6.8); bits 3:0 of that byte and bytes 129-131 of
that page are cleared. Such a store may cause a VM exit to occur after it
completes (see Section 22.1.3).
— If the “CR8-load exiting” VM-execution control is 1, MOV to CR8 causes a
VM exit (see Section 22.1.3); the “use TPR shadow” VM-execution control is
ignored in this case.
•
MWAIT. Behavior of the MWAIT instruction (which always causes an invalidopcode exception—#UD—if CPL > 0) is determined by the setting of the “MWAIT
exiting” VM-execution control:
— If the “MWAIT exiting” VM-execution control is 1, MWAIT causes a VM exit
(see Section 22.1.3).
— If the “MWAIT exiting” VM-execution control is 0, MWAIT operates normally if
any of the following is true: (1) the “interrupt-window exiting” VM-execution
control is 0; (2) ECX[0] is 0; or (3) RFLAGS.IF = 1.
— If the “MWAIT exiting” VM-execution control is 0, the “interrupt-window
exiting” VM-execution control is 1, ECX[0] = 1, and RFLAGS.IF = 0, MWAIT
does not cause the processor to enter an implementation-dependent
optimized state; instead, control passes to the instruction following the
MWAIT instruction.
•
RDMSR. Section 22.1.3 identifies when executions of the RDMSR instruction
cause VM exits. If such an execution causes neither a fault due to CPL > 0 nor a
VM exit, the instruction’s behavior may be modified for certain values of ECX:
— If ECX contains 10H (indicating the IA32_TIME_STAMP_COUNTER MSR), the
value returned by the instruction is determined by the setting of the “use TSC
offsetting” VM-execution control as well as the TSC offset:
•
If the control is 0, the instruction operates normally, loading EAX:EDX
with the value of the IA32_TIME_STAMP_COUNTER MSR.
•
If the control is 1, the instruction loads EAX:EDX with the sum (using
signed addition) of the value of the IA32_TIME_STAMP_COUNTER MSR
and the value of the TSC offset (interpreted as a signed value).
Vol. 3B 22-19
VMX NON-ROOT OPERATION
The 1-setting of the “use TSC-offsetting” VM-execution control does not
effect executions of RDMSR if ECX contains 6E0H (indicating the
IA32_TSC_DEADLINE MSR). Such executions return the APIC-timer deadline
relative to the actual timestamp counter without regard to the TSC offset.
— If ECX contains 808H (indicating the TPR MSR), instruction behavior is
determined by the setting of the “virtualize x2APIC mode” VM-execution
control:1
•
•
If the control is 0, the instruction operates normally. If the local APIC is in
x2APIC mode, EAX[7:0] is loaded with the value of the APIC’s taskpriority register (EDX and EAX[31:8] are cleared to 0). If the local APIC is
not in x2APIC mode, a general-protection fault occurs.
•
If the control is 1, the instruction loads EAX:EDX with the value of
bytes 87H:80H of the virtual-APIC page. This occurs even if the local APIC
is not in x2APIC mode (no general-protection fault occurs because the
local APIC is not x2APIC mode).
RDTSC. Behavior of the RDTSC instruction is determined by the settings of the
“RDTSC exiting” and “use TSC offsetting” VM-execution controls as well as the
TSC offset:
— If both controls are 0, RDTSC operates normally.
— If the “RDTSC exiting” VM-execution control is 0 and the “use TSC offsetting”
VM-execution control is 1, RDTSC loads EAX:EDX with the sum (using signed
addition) of the value of the IA32_TIME_STAMP_COUNTER MSR and the
value of the TSC offset (interpreted as a signed value).
— If the “RDTSC exiting” VM-execution control is 1, RDTSC causes a VM exit
(see Section 22.1.3).
•
RDTSCP. Behavior of the RDTSCP instruction is determined first by the setting of
the “enable RDTSCP” VM-execution control:2
— If the “enable RDTSCP” VM-execution control is 0, RDTSCP causes an invalidopcode exception (#UD).
— If the “enable RDTSCP” VM-execution control is 1, treatment is based on the
settings the “RDTSC exiting” and “use TSC offsetting” VM-execution controls
as well as the TSC offset:
•
If both controls are 0, RDTSCP operates normally.
1. “Virtualize x2APIC mode” is a secondary processor-based VM-execution control. If bit 31 of the
primary processor-based VM-execution controls is 0, VMX non-root operation functions as if the
“virtualize x2APIC mode” VM-execution control were 0. See Section 21.6.2.
2. “Enable RDTSCP” is a secondary processor-based VM-execution control. If bit 31 of the primary
processor-based VM-execution controls is 0, VMX non-root operation functions as if the “enable
RDTSCP” VM-execution control were 0. See Section 21.6.2.
22-20 Vol. 3B
VMX NON-ROOT OPERATION
•
•
If the “RDTSC exiting” VM-execution control is 0 and the “use TSC
offsetting” VM-execution control is 1, RDTSCP loads EAX:EDX with the
sum (using signed addition) of the value of the
IA32_TIME_STAMP_COUNTER MSR and the value of the TSC offset (interpreted as a signed value); it also loads ECX with the value of bits 31:0 of
the IA32_TSC_AUX MSR.
•
If the “RDTSC exiting” VM-execution control is 1, RDTSCP causes a
VM exit (see Section 22.1.3).
SMSW. The behavior of SMSW is determined by the CR0 guest/host mask and
the CR0 read shadow. For each position corresponding to a bit clear in the CR0
guest/host mask, the destination operand is loaded with the value of the corresponding bit in CR0. For each position corresponding to a bit set in the CR0
guest/host mask, the destination operand is loaded with the value of the corresponding bit in the CR0 read shadow. Thus, if every bit is cleared in the CR0
guest/host mask, MOV from CR0 reads normally from CR0; if every bit is set in
the CR0 guest/host mask, MOV from CR0 returns the value of the CR0 read
shadow.
Note the following: (1) for any memory destination or for a 16-bit register destination, only the low 16 bits of the CR0 guest/host mask and the CR0 read shadow
are used (bits 63:16 of a register destination are left unchanged); (2) for a 32-bit
register destination, only the low 32 bits of the CR0 guest/host mask and the CR0
read shadow are used (bits 63:32 of the destination are cleared); and
(3) depending on the contents of the CR0 guest/host mask and the CR0 read
shadow, bits may be set in the destination that would never be set when reading
directly from CR0.
•
WRMSR. Section 22.1.3 identifies when executions of the WRMSR instruction
cause VM exits. If such an execution neither a fault due to CPL > 0 nor a VM exit,
the instruction’s behavior may be modified for certain values of ECX:
— If ECX contains 79H (indicating IA32_BIOS_UPDT_TRIG MSR), no microcode
update is loaded, and control passes to the next instruction. This implies that
microcode updates cannot be loaded in VMX non-root operation.
— If ECX contains 808H (indicating the TPR MSR) and either EDX or EAX[31:8]
is non-zero, a general-protection fault occurs (this is true even if the logical
processor is not in VMX non-root operation). Otherwise, instruction behavior
is determined by the setting of the “virtualize x2APIC mode” VM-execution
control and the value of the TPR-threshold VM-execution control field:
•
If the control is 0, the instruction operates normally. If the local APIC is in
x2APIC mode, the value of EAX[7:0] is written to the APIC’s task-priority
register. If the local APIC is not in x2APIC mode, a general-protection
fault occurs.
•
If the control is 1, the instruction stores the value of EAX:EDX to
bytes 87H:80H of the virtual-APIC page. This store occurs even if the
local APIC is not in x2APIC mode (no general-protection fault occurs
Vol. 3B 22-21
VMX NON-ROOT OPERATION
because the local APIC is not x2APIC mode). The store may cause a
VM exit to occur after the instruction completes (see Section 22.1.3).
•
The 1-setting of the “use TSC-offsetting” VM-execution control does not
effect executions of WRMSR if ECX contains 10H (indicating the
IA32_TIME_STAMP_COUNTER MSR). Such executions modify the actual
timestamp counter without regard to the TSC offset.
•
The 1-setting of the “use TSC-offsetting” VM-execution control does not
effect executions of WRMSR if ECX contains 6E0H (indicating the
IA32_TSC_DEADLINE MSR). Such executions modify the APIC-timer
deadline relative to the actual timestamp counter without regard to the
TSC offset.
22.5
APIC ACCESSES THAT DO NOT CAUSE VM EXITS
As noted in Section 22.2, if the “virtualize APIC accesses” VM-execution control is 1,
most memory accesses to the APIC-access page (see Section 21.6.2) cause APICaccess VM exits.1 Section 22.2 identifies potential exceptions. These are covered in
Section 22.5.1 through Section 22.5.3.
In some cases, an attempt to access memory on the APIC-access page is converted
to an access to the virtual-APIC page (see Section 21.6.8). In these cases, the access
uses the memory type reported in bit 53:50 of the IA32_VMX_BASIC MSR (see
Appendix G.1).
22.5.1
Linear Accesses to the APIC-Access Page Using Large-Page
Translations
As noted in Section 22.2.1, a linear access to the APIC-access page using translation
with a large page (2-MByte, 4-MByte, or 1-GByte) may or may not cause an APICaccess VM exit. If it does not and the access is not a VTPR access (see Section
22.2.4), the access operates on memory on the APIC-access page. Section 22.5.3
describes the treatment if there is no APIC-access VM exit and the access is a VTPR
access.
22.5.2
Physical Accesses to the APIC-Access Page
A physical access to the APIC-access page may or may not cause an APIC-access
VM exit. If it does not and the access is not a VTPR access (see Section 22.2.4), the
access operates on memory on the APIC-access page (this may happen if the access
1. “Virtualize APIC accesses” is a secondary processor-based VM-execution control. If bit 31 of the
primary processor-based VM-execution controls is 0, VMX non-root operation functions as if the
“virtualize APIC accesses” VM-execution control were 0. See Section 21.6.2.
22-22 Vol. 3B
VMX NON-ROOT OPERATION
causes an APIC-access VM exit). Section 22.5.3 describes the treatment if there is no
APIC-access VM exit and the access is a VTPR access.
22.5.3
VTPR Accesses
As noted in Section 22.2.4, a memory access is a VTPR access if all of the following
hold: (1) the “use TPR shadow” VM-execution control is 1; (2) the access is not for
an instruction fetch; (3) the access is at most 32 bits in width; and (4) the access is
to offset 80H on the APIC-access page.
The treatment of VTPR accesses depends on the nature of the access:
•
A linear VTPR access using a translation with a 4-KByte page does not cause an
APIC-access VM exit. Instead, it is converted so that, instead of accessing offset
80H on the APIC-access page, it accesses offset 80H on the virtual-APIC page.
Further details are provided in Section 22.5.3.1 to Section 22.5.3.3.
•
A linear VTPR access using a translation with a large page (2-MByte, 4-MByte, or
1-GByte) may be treated in either of two ways:
— It may operate on memory on the APIC-access page. The details in Section
22.5.3.1 to Section 22.5.3.3 do not apply.
— It may be converted so that, instead of accessing offset 80H on the APICaccess page, it accesses offset 80H on the virtual-APIC page. Further details
are provided in Section 22.5.3.1 to Section 22.5.3.3.
•
A physical VTPR access may be treated in one of three ways:
— It may cause an APIC-access VM exit. The details in Section 22.5.3.1 to
Section 22.5.3.3 do not apply.
— It may operate on memory on the APIC-access page (and possibly then cause
an APIC-access VM exit). The details in Section 22.5.3.1 to Section 22.5.3.3
do not apply.
— It may be converted so that, instead of accessing offset 80H on the APICaccess page, it accesses offset 80H on the virtual-APIC page. Further details
are provided in Section 22.5.3.1 to Section 22.5.3.3.
Linear VTPR accesses never cause APIC-access VM exits (recall that an access is a
VTPR access only if the “use TPR shadow” VM-execution control is 1).
22.5.3.1
Treatment of Individual VTPR Accesses
The following items detail the treatment of VTPR accesses:
•
VTPR read accesses. Such an access completes normally (reading data from the
field at offset 80H on the virtual-APIC page).
The following items detail certain instructions that are considered to perform
read accesses and how they behavior when accessing the VTPR:
Vol. 3B 22-23
VMX NON-ROOT OPERATION
— A VTPR access using the CLFLUSH instruction flushes data for offset 80H on
the virtual-APIC page.
— A VTPR access using the LMSW instruction may cause a VM exit due to the
CR0 guest/host mask and the CR0 read shadow.
— A VTPR access using the MONITOR instruction causes the logical processor to
monitor offset 80H on the virtual-APIC page.
— A VTPR access using the PREFETCH instruction may prefetch data; if so, it is
from offset 80H on the virtual-APIC page.
•
VTPR write accesses. Such an access completes normally (writing data to the
field at offset 80H on the virtual-APIC page) and causes a TPR-shadow update
(see Section 22.5.3.3).
The following items detail certain instructions that are considered to perform
write accesses and how they behavior when accessing the VTPR:
— The ENTER instruction is considered to write to VTPR if the byte referenced by
the final value of the stack pointer is at offset 80H on the APIC-access page
(even though ENTER does not write to that byte if its size operand is nonzero). The instruction is followed by a TPR-shadow update.
— A VTPR access using the SMSW instruction stores data determined by the
current CR0 contents, the CR0 guest/host mask, and the CR0 read shadow.
The instruction is followed by a TPR-shadow update.
22.5.3.2
Operations with Multiple Accesses
Some operations may access multiple addresses. These operations include the
execution of some instructions and the delivery of events through the IDT (including
those injected with VM entry). In some cases, the Intel® 64 architecture specifies the
ordering of these memory accesses. The following items describe the treatment of
VTPR accesses that are part of such multi-access operations:
•
Read-modify-write instructions may first perform a VTPR read access and then a
VTPR write access. Both accesses complete normally (as described in Section
22.5.3.1). The instruction is followed by a TPR-shadow update (see Section
22.5.3.3).
•
Some operations may perform a VTPR write access and subsequently cause a
fault. This situation is treated as follows:
— If the fault leads to a VM exit, no TPR-shadow update occurs.
— If the fault does not lead to a VM exit, a TPR-shadow update occurs after fault
delivery completes and before execution of the fault handler.
•
If an operation includes a VTPR access and an access to some other field on the
APIC-access page, the latter access causes an APIC-access VM exit as described
in Section 22.2.
If the operation performs a VTPR write access before the APIC-access VM exit,
there is no TPR-shadow update.
22-24 Vol. 3B
VMX NON-ROOT OPERATION
•
Suppose that the first iteration of a repeated string instruction (including OUTS)
that accesses the APIC-access page performs a VTPR read access and that the
next iteration would read from the APIC-access page using an offset other than
80H. The following items describe the behavior of the logical processor:
— The iteration that performs the VTPR read access completes successfully,
reading data from offset 80H on the virtual-APIC page.
— The iteration that would read from the other offset causes an APIC-access
VM exit. The instruction pointer saved in the VMCS references the repeated
string instruction and the values of the general-purpose registers are such
that iteration would be repeated if the instruction were restarted.
•
Suppose that the first iteration of a repeated string instruction (including INS)
that accesses the APIC-access page performs a VTPR write access and that the
next iteration would write to the APIC-access page using an offset other than
80H. The following items describe the behavior of the logical processor:
— The iteration that performs the VTPR write access writes data to offset 80H on
the virtual-APIC page. The write is followed by a TPR-shadow update, which
may cause a VM exit (see Section 22.5.3.3).
— If the TPR-shadow update does cause a VM exit, the instruction pointer saved
in the VMCS references the repeated string instruction and the values of the
general-purpose registers are such that the next iteration would be
performed if the instruction were restarted.
— If the TPR-shadow update does not cause a VM exit, the iteration that would
write to the other offset causes an APIC-access VM exit. The instruction
pointer saved in the VMCS references the repeated string instruction and the
values of the general-purpose registers are such that that iteration would be
repeated if the instruction were restarted.
•
Suppose that the last iteration of a repeated string instruction (including INS)
performs a VTPR write access. The iteration writes data to offset 80H on the
virtual-APIC page. The write is followed by a TPR-shadow update, which may
cause a VM exit (see Section 22.5.3.3). If it does, the instruction pointer saved in
the VMCS references the instruction after the string instruction and the values of
the general-purpose registers reflect completion of the string instruction.
22.5.3.3
TPR-Shadow Updates
If the “use TPR shadow” and “virtualize APIC accesses” VM-execution controls are
both 1, a logical processor performs certain actions after any operation (or iteration
of a repeated string instruction) with a VTPR write access. These actions are called a
TPR-shadow update. (As noted in Section 22.5.3.2, a TPR-shadow update does not
occur following an access that causes a VM exit.)
A TPR-shadow update includes the following actions:
1. Bits 31:8 at offset 80H on the virtual-APIC page are cleared.
Vol. 3B 22-25
VMX NON-ROOT OPERATION
2. If the value of bits 3:0 of the TPR threshold VM-execution control field is greater
than the value of bits 7:4 at offset 80H on the virtual-APIC page, a VM exit will
occur.
TPR-shadow updates take priority over system-management interrupts (SMIs), INIT
signals, and lower priority events. A TPR-shadow update thus has priority over any
debug exceptions that may have been triggered by the operation causing the TPRshadow update. TPR-shadow updates (and any VM exits they cause) are not blocked
if RFLAGS.IF = 0 or by the MOV SS, POP SS, or STI instructions.
22.6
OTHER CHANGES IN VMX NON-ROOT OPERATION
Treatments of event blocking and of task switches differ in VMX non-root operation as
described in the following sections.
22.6.1
Event Blocking
Event blocking is modified in VMX non-root operation as follows:
•
If the “external-interrupt exiting” VM-execution control is 1, RFLAGS.IF does not
control the blocking of external interrupts. In this case, an external interrupt that
is not blocked for other reasons causes a VM exit (even if RFLAGS.IF = 0).
•
If the “external-interrupt exiting” VM-execution control is 1, external interrupts
may or may not be blocked by STI or by MOV SS (behavior is implementationspecific).
•
If the “NMI exiting” VM-execution control is 1, non-maskable interrupts (NMIs)
may or may not be blocked by STI or by MOV SS (behavior is implementationspecific).
22.6.2
Treatment of Task Switches
Task switches are not allowed in VMX non-root operation. Any attempt to effect a
task switch in VMX non-root operation causes a VM exit. However, the following
checks are performed (in the order indicated), possibly resulting in a fault, before
there is any possibility of a VM exit due to task switch:
1. If a task gate is being used, appropriate checks are made on its P bit and on the
proper values of the relevant privilege fields. The following cases detail the
privilege checks performed:
a. If CALL, INT n, or JMP accesses a task gate in IA-32e mode, a generalprotection exception occurs.
b. If CALL, INT n, INT3, INTO, or JMP accesses a task gate outside IA-32e mode,
privilege-levels checks are performed on the task gate but, if they pass,
22-26 Vol. 3B
VMX NON-ROOT OPERATION
privilege levels are not checked on the referenced task-state segment (TSS)
descriptor.
c.
If CALL or JMP accesses a TSS descriptor directly in IA-32e mode, a generalprotection exception occurs.
d. If CALL or JMP accesses a TSS descriptor directly outside IA-32e mode,
privilege levels are checked on the TSS descriptor.
e. If a non-maskable interrupt (NMI), an exception, or an external interrupt
accesses a task gate in the IDT in IA-32e mode, a general-protection
exception occurs.
f.
If a non-maskable interrupt (NMI), an exception other than breakpoint
exceptions (#BP) and overflow exceptions (#OF), or an external interrupt
accesses a task gate in the IDT outside IA-32e mode, no privilege checks are
performed.
g. If IRET is executed with RFLAGS.NT = 1 in IA-32e mode, a generalprotection exception occurs.
h. If IRET is executed with RFLAGS.NT = 1 outside IA-32e mode, a TSS
descriptor is accessed directly and no privilege checks are made.
2. Checks are made on the new TSS selector (for example, that is within GDT
limits).
3. The new TSS descriptor is read. (A page fault results if a relevant GDT page is not
present).
4. The TSS descriptor is checked for proper values of type (depends on type of task
switch), P bit, S bit, and limit.
Only if checks 1–4 all pass (do not generate faults) might a VM exit occur. However,
the ordering between a VM exit due to a task switch and a page fault resulting from
accessing the old TSS or the new TSS is implementation-specific. Some logical
processors may generate a page fault (instead of a VM exit due to a task switch) if
accessing either TSS would cause a page fault. Other logical processors may
generate a VM exit due to a task switch even if accessing either TSS would cause a
page fault.
If an attempt at a task switch through a task gate in the IDT causes an exception
(before generating a VM exit due to the task switch) and that exception causes a
VM exit, information about the event whose delivery that accessed the task gate is
recorded in the IDT-vectoring information fields and information about the exception
that caused the VM exit is recorded in the VM-exit interruption-information fields.
See Section 24.2. The fact that a task gate was being accessed is not recorded in the
VMCS.
If an attempt at a task switch through a task gate in the IDT causes VM exit due to
the task switch, information about the event whose delivery accessed the task gate
is recorded in the IDT-vectoring fields of the VMCS. Since the cause of such a VM exit
is a task switch and not an interruption, the valid bit for the VM-exit interruption
information field is 0. See Section 24.2.
Vol. 3B 22-27
VMX NON-ROOT OPERATION
22.7
FEATURES SPECIFIC TO VMX NON-ROOT OPERATION
Some VM-execution controls cause VM exits using features that are specific to VMX
non-root operation. These are the VMX-preemption timer (Section 22.7.1) and the
monitor trap flag (Section 22.7.2).
22.7.1
VMX-Preemption Timer
If the last VM entry was performed with the 1-setting of “activate VMX-preemption
timer” VM-execution control, the VMX-preemption timer counts down (from the
value loaded by VM entry; see Section 23.6.4) in VMX non-root operation. When the
timer counts down to zero, it stops counting down and a VM exit occurs (see Section
22.3).
The VMX-preemption timer counts down at rate proportional to that of the timestamp
counter (TSC). Specifically, the timer counts down by 1 every time bit X in the TSC
changes due to a TSC increment. The value of X is in the range 0–31 and can be
determined by consulting the VMX capability MSR IA32_VMX_MISC (see Appendix
G.6).
The VMX-preemption timer operates in the C-states C0, C1, and C2; it also operates
in the shutdown and wait-for-SIPI states. If the timer counts down to zero in C1, C2,
or shutdown, the logical processor transitions to the C0 C-state and causes a VM exit.
(The timer does not cause a VM exit if it counts down to zero in the wait-for-SIPI
state.) The timer is not decremented and does not cause VM exits in C-states deeper
than C2.
Treatment of the timer in the case of system management interrupts (SMIs) and
system-management mode (SMM) depends on whether the treatment of SMIs and
SMM:
•
If the default treatment of SMIs and SMM (see Section 26.14) is active, the VMXpreemption timer counts across an SMI to VMX non-root operation, subsequent
execution in SMM, and the return from SMM via the RSM instruction. However,
the timer can cause a VM exit only from VMX non-root operation. If the timer
expires during SMI, in SMM, or during RSM, a timer-induced VM exit occurs
immediately after RSM with its normal priority unless it is blocked based on
activity state (Section 22.3).
•
If the dual-monitor treatment of SMIs and SMM (see Section 26.15) is active,
transitions into and out of SMM are VM exits and VM entries, respectively. The
treatment of the VMX-preemption timer by those transitions is mostly the same
as for ordinary VM exits and VM entries; Section 26.15.2 and Section 26.15.4
detail some differences.
22-28 Vol. 3B
VMX NON-ROOT OPERATION
22.7.2
Monitor Trap Flag
The monitor trap flag is a debugging feature that causes VM exits to occur on
certain instruction boundaries in VMX non-root operation. Such VM exits are called
MTF VM exits. An MTF VM exit may occur on an instruction boundary in VMX nonroot operation as follows:
•
If the “monitor trap flag” VM-execution control is 1 and VM entry is injecting a
vectored event (see Section 23.5.1), an MTF VM exit is pending on the instruction
boundary before the first instruction following the VM entry.
•
If VM entry is injecting a pending MTF VM exit (see Section 23.5.2), an MTF
VM exit is pending on the instruction boundary before the first instruction
following the VM entry. This is the case even if the “monitor trap flag” VMexecution control is 0.
•
If the “monitor trap flag” VM-execution control is 1, VM entry is not injecting an
event, and a pending event (e.g., debug exception or interrupt) is delivered
before an instruction can execute, an MTF VM exit is pending on the instruction
boundary following delivery of the event (or any nested exception).
•
Suppose that the “monitor trap flag” VM-execution control is 1, VM entry is not
injecting an event, and the first instruction following VM entry is a REP-prefixed
string instruction:
— If the first iteration of the instruction causes a fault, an MTF VM exit is
pending on the instruction boundary following delivery of the fault (or any
nested exception).
— If the first iteration of the instruction does not cause a fault, an MTF VM exit
is pending on the instruction boundary after that iteration.
•
Suppose that the “monitor trap flag” VM-execution control is 1, VM entry is not
injecting an event, and the first instruction following VM entry is not a REPprefixed string instruction:
— If the instruction causes a fault, an MTF VM exit is pending on the instruction
boundary following delivery of the fault (or any nested exception).1
— If the instruction does not cause a fault, an MTF VM exit is pending on the
instruction boundary following execution of that instruction. If the instruction
is INT3 or INTO, this boundary follows delivery of any software exception. If
the instruction is INT n, this boundary follows delivery of a software interrupt.
If the instruction is HLT, the MTF VM exit will be from the HLT activity state.
No MTF VM exit occurs if another VM exit occurs before reaching the instruction
boundary on which an MTF VM exit would be pending (e.g., due to an exception or
triple fault).
1. This item includes the cases of an invalid opcode exception—#UD— generated by the UD2
instruction and a BOUND-range exceeded exception—#BR—generated by the BOUND instruction.
Vol. 3B 22-29
VMX NON-ROOT OPERATION
An MTF VM exit occurs on the instruction boundary on which it is pending unless a
higher priority event takes precedence or the MTF VM exit is blocked due to the
activity state:
•
System-management interrupts (SMIs), INIT signals, and higher priority events
take priority over MTF VM exits. MTF VM exits take priority over debug-trap
exceptions and lower priority events.
•
No MTF VM exit occurs if the processor is in either the shutdown activity state or
wait-for-SIPI activity state. If a non-maskable interrupt subsequently takes the
logical processor out of the shutdown activity state without causing a VM exit, an
MTF VM exit is pending after delivery of that interrupt.
22.7.3
Translation of Guest-Physical Addresses Using EPT
The extended page-table mechanism (EPT) is a feature that can be used to support
the virtualization of physical memory. When EPT is in use, certain physical addresses
are treated as guest-physical addresses and are not used to access memory directly.
Instead, guest-physical addresses are translated by traversing a set of EPT paging
structures to produce physical addresses that are used to access memory.
Details of the EPT are given in Section 25.2.
22.8
UNRESTRICTED GUESTS
The first processors to support VMX operation require CR0.PE and CR0.PG to be 1 in
VMX operation (see Section 20.8). This restriction implies that guest software cannot
be run in unpaged protected mode or in real-address mode. Later processors support
a VM-execution control called “unrestricted guest”.1 If this control is 1, CR0.PE and
CR0.PG may be 0 in VMX non-root operation. Such processors allow guest software
to run in unpaged protected mode or in real-address mode. The following items
describe the behavior of such software:
•
The MOV CR0 instructions does not cause a general-protection exception simply
because it would set either CR0.PE and CR0.PG to 0. See Section 22.4 for details.
•
A logical processor treats the values of CR0.PE and CR0.PG in VMX non-root
operation just as it does outside VMX operation. Thus, if CR0.PE = 0, the
processor operates as it does normally in real-address mode (for example, it uses
the 16-bit interrupt table to deliver interrupts and exceptions). If CR0.PG = 0,
the processor operates as it does normally when paging is disabled.
•
Processor operation is modified by the fact that the processor is in VMX non-root
operation and by the settings of the VM-execution controls just as it is in
1. “Unrestricted guest” is a secondary processor-based VM-execution control. If bit 31 of the primary processor-based VM-execution controls is 0, VMX non-root operation functions as if the
“unrestricted guest” VM-execution control were 0. See Section 21.6.2.
22-30 Vol. 3B
VMX NON-ROOT OPERATION
protected mode or when paging is enabled. Instructions, interrupts, and
exceptions that cause VM exits in protected mode or when paging is enabled also
do so in real-address mode or when paging is disabled. The following examples
should be noted:
— If CR0.PG = 0, page faults do not occur and thus cannot cause VM exits.
— If CR0.PE = 0, invalid-TSS exceptions do not occur and thus cannot cause
VM exits.
— If CR0.PE = 0, the following instructions cause invalid-opcode exceptions and
do not cause VM exits: INVEPT, INVVPID, LLDT, LTR, SLDT, STR, VMCLEAR,
VMLAUNCH, VMPTRLD, VMPTRST, VMREAD, VMRESUME, VMWRITE, VMXOFF,
and VMXON.
•
If CR0.PG = 0, each linear address is passed directly to the EPT mechanism for
translation to a physical address.1 The guest memory type passed on to the EPT
mechanism is WB (writeback).
1. As noted in Section 23.2.1.1, the “enable EPT” VM-execution control must be 1 if the “unrestricted guest” VM-execution control is 1.
Vol. 3B 22-31
VMX NON-ROOT OPERATION
22-32 Vol. 3B
CHAPTER 23
VM ENTRIES
Software can enter VMX non-root operation using either of the VM-entry instructions
VMLAUNCH and VMRESUME. VMLAUNCH can be used only with a VMCS whose launch
state is clear and VMRESUME can be used only with a VMCS whose the launch state
is launched. VMLAUNCH should be used for the first VM entry after VMCLEAR; VMRESUME should be used for subsequent VM entries with the same VMCS.
Each VM entry performs the following steps in the order indicated:
1. Basic checks are performed to ensure that VM entry can commence
(Section 23.1).
2. The control and host-state areas of the VMCS are checked to ensure that they are
proper for supporting VMX non-root operation and that the VMCS is correctly
configured to support the next VM exit (Section 23.2).
3. The following may be performed in parallel or in any order (Section 23.3):
•
The guest-state area of the VMCS is checked to ensure that, after the
VM entry completes, the state of the logical processor is consistent with
IA-32 and Intel 64 architectures.
•
Processor state is loaded from the guest-state area and based on controls in
the VMCS.
•
Address-range monitoring is cleared.
4. MSRs are loaded from the VM-entry MSR-load area (Section 23.4).
5. If VMLAUNCH is being executed, the launch state of the VMCS is set to
“launched.”
6. An event may be injected in the guest context (Section 23.5).
Steps 1–4 above perform checks that may cause VM entry to fail. Such failures occur
in one of the following three ways:
•
Some of the checks in Section 23.1 may generate ordinary faults (for example,
an invalid-opcode exception). Such faults are delivered normally.
•
Some of the checks in Section 23.1 and all the checks in Section 23.2 cause
control to pass to the instruction following the VM-entry instruction. The failure is
indicated by setting RFLAGS.ZF1 (if there is a current VMCS) or RFLAGS.CF (if
there is no current VMCS). If there is a current VMCS, an error number indicating
the cause of the failure is stored in the VM-instruction error field. See Chapter 5
1. This chapter uses the notation RAX, RIP, RSP, RFLAGS, etc. for processor registers because most
processors that support VMX operation also support Intel 64 architecture. For IA-32 processors,
this notation refers to the 32-bit forms of those registers (EAX, EIP, ESP, EFLAGS, etc.). In a few
places, notation such as EAX is used to refer specifically to lower 32 bits of the indicated register.
Vol. 3B 23-1
VM ENTRIES
of the Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume
2B for the error numbers.
•
The checks in Section 23.3 and Section 23.4 cause processor state to be loaded
from the host-state area of the VMCS (as would be done on a VM exit).
Information about the failure is stored in the VM-exit information fields. See
Section 23.7 for details.
EFLAGS.TF = 1 causes a VM-entry instruction to generate a single-step debug exception only if failure of one of the checks in Section 23.1 and Section 23.2 causes
control to pass to the following instruction. A VM-entry does not generate a singlestep debug exception in any of the following cases: (1) the instruction generates a
fault; (2) failure of one of the checks in Section 23.3 or in loading MSRs causes
processor state to be loaded from the host-state area of the VMCS; or (3) the instruction passes all checks in Section 23.1, Section 23.2, and Section 23.3 and there is no
failure in loading MSRs.
Section 26.15 describes the dual-monitor treatment of system-management interrupts (SMIs) and system-management mode (SMM). Under this treatment, code
running in SMM returns using VM entries instead of the RSM instruction. A VM entry
returns from SMM if it is executed in SMM and the “entry to SMM” VM-entry control
is 0. VM entries that return from SMM differ from ordinary VM entries in ways that
are detailed in Section 26.15.4.
23.1
BASIC VM-ENTRY CHECKS
Before a VM entry commences, the current state of the logical processor is checked
in the following order:
1. If the logical processor is in virtual-8086 mode or compatibility mode, an
invalid-opcode exception is generated.
2. If the current privilege level (CPL) is not zero, a general-protection exception is
generated.
3. If there is no current VMCS, RFLAGS.CF is set to 1 and control passes to the next
instruction.
4. If there is a current VMCS, the following conditions are evaluated in order; any of
these cause VM entry to fail:
a. if there is MOV-SS blocking (see Table 21-3)
b. if the VM entry is invoked by VMLAUNCH and the VMCS launch state is not
clear
c.
if the VM entry is invoked by VMRESUME and the VMCS launch state is not
launched
If any of these checks fail, RFLAGS.ZF is set to 1 and control passes to the next
instruction. An error number indicating the cause of the failure is stored in the
23-2 Vol. 3B
VM ENTRIES
VM-instruction error field. See Chapter 5 of the Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 2B for the error numbers.
23.2
CHECKS ON VMX CONTROLS AND HOST-STATE AREA
If the checks in Section 23.1 do not cause VM entry to fail, the control and host-state
areas of the VMCS are checked to ensure that they are proper for supporting VMX
non-root operation, that the VMCS is correctly configured to support the next
VM exit, and that, after the next VM exit, the processor’s state is consistent with the
Intel 64 and IA-32 architectures.
VM entry fails if any of these checks fail. When such failures occur, control is passed
to the next instruction, RFLAGS.ZF is set to 1 to indicate the failure, and the
VM-instruction error field is loaded with an error number that indicates whether the
failure was due to the controls or the host-state area (see Chapter 5 of the Intel® 64
and IA-32 Architectures Software Developer’s Manual, Volume 2B).
These checks may be performed in any order. Thus, an indication by error number of
one cause (for example, host state) does not imply that there are not also other
errors. Different processors may thus give different error numbers for the same
VMCS. Some checks prevent establishment of settings (or combinations of settings)
that are currently reserved. Future processors may allow such settings (or combinations) and may not perform the corresponding checks. The correctness of software
should not rely on VM-entry failures resulting from the checks documented in this
section.
The checks on the controls and the host-state area are presented in Section 23.2.1
through Section 23.2.4. These sections reference VMCS fields that correspond to
processor state. Unless otherwise stated, these references are to fields in the hoststate area.
23.2.1
Checks on VMX Controls
This section identifies VM-entry checks on the VMX control fields.
23.2.1.1
VM-Execution Control Fields
VM entries perform the following checks on the VM-execution control fields:1
•
Reserved bits in the pin-based VM-execution controls must be set properly.
Software may consult the VMX capability MSRs to determine the proper settings
(see Appendix G.3.1).
1. If the “activate secondary controls” primary processor-based VM-execution control is 0, VM entry
operates as if each secondary processor-based VM-execution control were 0.
Vol. 3B 23-3
VM ENTRIES
•
Reserved bits in the primary processor-based VM-execution controls must be set
properly. Software may consult the VMX capability MSRs to determine the proper
settings (see Appendix G.3.2).
•
If the “activate secondary controls” primary processor-based VM-execution
control is 1, reserved bits in the secondary processor-based VM-execution
controls must be set properly. Software may consult the VMX capability MSRs to
determine the proper settings (see Appendix G.3.3).
If the “activate secondary controls” primary processor-based VM-execution
control is 0 (or if the processor does not support the 1-setting of that control),
no checks are performed on the secondary processor-based VM-execution
controls. The logical processor operates as if all the secondary processor-based
VM-execution controls were 0.
•
The CR3-target count must not be greater than 4. Future processors may support
a different number of CR3-target values. Software should read the VMX
capability MSR IA32_VMX_MISC to determine the number of values supported
(see Appendix G.6).
•
If the “use I/O bitmaps” VM-execution control is 1, bits 11:0 of each I/O-bitmap
address must be 0. Neither address should set any bits beyond the processor’s
physical-address width.1,2
•
If the “use MSR bitmaps” VM-execution control is 1, bits 11:0 of the MSR-bitmap
address must be 0. The address should not set any bits beyond the processor’s
physical-address width.3
•
If the “use TPR shadow” VM-execution control is 1, the virtual-APIC address must
satisfy the following checks:
— Bits 11:0 of the address must be 0.
— The address should not set any bits beyond the processor’s physical-address
width.4
The following items describe the treatment of bytes 81H-83H on the virtualAPIC page (see Section 21.6.8) if all of the above checks are satisfied and the
“use TPR shadow” VM-execution control is 1, treatment depends upon the
setting of the “virtualize APIC accesses” VM-execution control:5
— If the “virtualize APIC accesses” VM-execution control is 0, the bytes may be
cleared. (If the bytes are not cleared, they are left unmodified.)
1. Software can determine a processor’s physical-address width by executing CPUID with
80000008H in EAX. The physical-address width is returned in bits 7:0 of EAX.
2. If IA32_VMX_BASIC[48] is read as 1, these addresses must not set any bits in the range 63:32;
see Appendix G.1.
3. If IA32_VMX_BASIC[48] is read as 1, this address must not set any bits in the range 63:32; see
Appendix G.1.
4. If IA32_VMX_BASIC[48] is read as 1, this address must not set any bits in the range 63:32; see
Appendix G.1.
23-4 Vol. 3B
VM ENTRIES
— If the “virtualize APIC accesses” VM-execution control is 1, the bytes are
cleared.
— If the VM entry fails, the any clearing of the bytes may or may not occur. This
is true either if the failure causes control to pass to the instruction following
the VM-entry instruction or if it cause processor state to be loaded from the
host-state area of the VMCS. Behavior may be implementation-specific.
•
If the “use TPR shadow” VM-execution control is 1, bits 31:4 of the TPR threshold
VM-execution control field must be 0.
•
The following check is performed if the “use TPR shadow” VM-execution control is
1 and the “virtualize APIC accesses” VM-execution control is 0: the value of
bits 3:0 of the TPR threshold VM-execution control field should not be greater
than the value of bits 7:4 in byte 80H on the virtual-APIC page (see Section
21.6.8).
•
If the “NMI exiting” VM-execution control is 0, the “virtual NMIs” VM-execution
control must be 0.
•
If the “virtual NMIs” VM-execution control is 0, the “NMI-window exiting” VMexecution control must be 0.
•
If the “virtualize APIC-accesses” VM-execution control is 1, the APIC-access
address must satisfy the following checks:
— Bits 11:0 of the address must be 0.
— The address should not set any bits beyond the processor’s physical-address
width.1
•
If the “virtualize x2APIC mode” VM-execution control is 1, the “use TPR shadow”
VM-execution control must be 1 and the “virtualize APIC accesses” VM-execution
control must be 0.2
•
If the “enable VPID” VM-execution control is 1, the value of the VPID VMexecution control field must not be 0000H.
•
If the “enable EPT” VM-execution control is 1, the EPTP VM-execution control field
(see Table 21-8 in Section 21.6.11) must satisfy the following checks:3
5. “Virtualize APIC accesses” is a secondary processor-based VM-execution control. If bit 31 of the
primary processor-based VM-execution controls is 0, VM entry functions as if the “virtualize APIC
accesses” VM-execution control were 0. See Section 21.6.2.
1. If IA32_VMX_BASIC[48] is read as 1, this address must not set any bits in the range 63:32; see
Appendix G.1.
2. “Virtualize APIC accesses” and “virtualize x2APIC mode” are both secondary processor-based VMexecution controls. If bit 31 of the primary processor-based VM-execution controls is 0, VM entry
functions as if both these controls were 0. See Section 21.6.2.
3. “Enable EPT” is a secondary processor-based VM-execution control. If bit 31 of the primary processor-based VM-execution controls is 0, VM entry functions as if the “enable EPT” VM-execution control were 0. See Section 21.6.2.
Vol. 3B 23-5
VM ENTRIES
— The EPT memory type (bits 2:0) must be a value supported by the logical
processor as indicated in the IA32_VMX_EPT_VPID_CAP MSR (see Appendix
G.10).
— Bits 5:3 (1 less than the EPT page-walk length) must be 3, indicating an EPT
page-walk length of 4; see Section 25.2.2.
— Reserved bits 11:6 and 63:N (where N is the processor’s physical-address
width) must all be 0.
— If the “unrestricted guest” VM-execution control is 1, the “enable EPT” VMexecution control must also be 1.1
23.2.1.2
VM-Exit Control Fields
VM entries perform the following checks on the VM-exit control fields.
•
Reserved bits in the VM-exit controls must be set properly. Software may consult
the VMX capability MSRs to determine the proper settings (see Appendix G.4).
•
If “activate VMX-preemption timer” VM-execution control is 0, the “save VMXpreemption timer value” VM-exit control must also be 0.
•
The following checks are performed for the VM-exit MSR-store address if the
VM-exit MSR-store count field is non-zero:
— The lower 4 bits of the VM-exit MSR-store address must be 0. The address
should not set any bits beyond the processor’s physical-address width.2
— The address of the last byte in the VM-exit MSR-store area should not set any
bits beyond the processor’s physical-address width. The address of this last
byte is VM-exit MSR-store address + (MSR count * 16) – 1. (The arithmetic
used for the computation uses more bits than the processor’s physicaladdress width.)
If IA32_VMX_BASIC[48] is read as 1, neither address should set any bits in the
range 63:32; see Appendix G.1.
•
The following checks are performed for the VM-exit MSR-load address if the
VM-exit MSR-load count field is non-zero:
— The lower 4 bits of the VM-exit MSR-load address must be 0. The address
should not set any bits beyond the processor’s physical-address width.
— The address of the last byte in the VM-exit MSR-load area should not set any
bits beyond the processor’s physical-address width. The address of this last
byte is VM-exit MSR-load address + (MSR count * 16) – 1. (The arithmetic
1. “Unrestricted guest” and “enable EPT” are both secondary processor-based VM-execution controls. If bit 31 of the primary processor-based VM-execution controls is 0, VM entry functions as
if both these controls were 0. See Section 21.6.2.
2. Software can determine a processor’s physical-address width by executing CPUID with
80000008H in EAX. The physical-address width is returned in bits 7:0 of EAX.
23-6 Vol. 3B
VM ENTRIES
used for the computation uses more bits than the processor’s physicaladdress width.)
If IA32_VMX_BASIC[48] is read as 1, neither address should set any bits in the
range 63:32; see Appendix G.1.
23.2.1.3
VM-Entry Control Fields
VM entries perform the following checks on the VM-entry control fields.
•
Reserved bits in the VM-entry controls must be set properly. Software may
consult the VMX capability MSRs to determine the proper settings (see Appendix
G.5).
•
Fields relevant to VM-entry event injection must be set properly. These fields are
the VM-entry interruption-information field (see Table 21-12 in Section 21.8.3),
the VM-entry exception error code, and the VM-entry instruction length. If the
valid bit (bit 31) in the VM-entry interruption-information field is 1, the following
must hold:
— The field’s interruption type (bits 10:8) is not set to a reserved value. Value 1
is reserved on all logical processors; value 7 (other event) is reserved on
logical processors that do not support the 1-setting of the “monitor trap flag”
VM-execution control.
— The field’s vector (bits 7:0) is consistent with the interruption type:
•
•
•
If the interruption type is non-maskable interrupt (NMI), the vector is 2.
If the interruption type is hardware exception, the vector is at most 31.
If the interruption type is other event, the vector is 0 (pending MTF
VM exit).
— The field's deliver-error-code bit (bit 11) is 1 if and only if (1) either (a) the
"unrestricted guest" VM-execution control is 0; or (b) bit 0 (corresponding to
CR0.PE) is set in the CR0 field in the guest-state area; (2) the interruption
type is hardware exception; and (3) the vector indicates an exception that
would normally deliver an error code (8 = #DF; 10 = TS; 11 = #NP; 12 =
#SS; 13 = #GP; 14 = #PF; or 17 = #AC).
— Reserved bits in the field (30:12) are 0.
— If the deliver-error-code bit (bit 11) is 1, bits 31:15 of the VM-entry
exception error-code field are 0.
— If the interruption type is software interrupt, software exception, or
privileged software exception, the VM-entry instruction-length field is in the
range 1–15.
•
The following checks are performed for the VM-entry MSR-load address if the
VM-entry MSR-load count field is non-zero:
— The lower 4 bits of the VM-entry MSR-load address must be 0. The address
should not set any bits beyond the processor’s physical-address width.1
Vol. 3B 23-7
VM ENTRIES
— The address of the last byte in the VM-entry MSR-load area should not set any
bits beyond the processor’s physical-address width. The address of this last
byte is VM-entry MSR-load address + (MSR count * 16) – 1. (The arithmetic
used for the computation uses more bits than the processor’s physicaladdress width.)
If IA32_VMX_BASIC[48] is read as 1, neither address should set any bits in the
range 63:32; see Appendix G.1.
•
If the processor is not in SMM, the “entry to SMM” and “deactivate dual-monitor
treatment” VM-entry controls must be 0.
•
The “entry to SMM” and “deactivate dual-monitor treatment” VM-entry controls
cannot both be 1.
23.2.2
Checks on Host Control Registers and MSRs
The following checks are performed on fields in the host-state area that correspond
to control registers and MSRs:
•
The CR0 field must not set any bit to a value not supported in VMX operation (see
Section 20.8).1
•
The CR4 field must not set any bit to a value not supported in VMX operation (see
Section 20.8).
•
On processors that support Intel 64 architecture, the CR3 field must be such that
bits 63:52 and bits in the range 51:32 beyond the processor’s physical-address
width must be 0.2,3
•
On processors that support Intel 64 architecture, the IA32_SYSENTER_ESP field
and the IA32_SYSENTER_EIP field must each contain a canonical address.
•
If the “load IA32_PERF_GLOBAL_CTRL” VM-exit control is 1, bits reserved in the
IA32_PERF_GLOBAL_CTRL MSR must be 0 in the field for that register (see
Figure 30-3).
•
If the “load IA32_PAT” VM-exit control is 1, the value of the field for the IA32_PAT
MSR must be one that could be written by WRMSR without fault at CPL 0. Specifically, each of the 8 bytes in the field must have one of the values 0 (UC), 1 (WC),
4 (WT), 5 (WP), 6 (WB), or 7 (UC-).
1. Software can determine a processor’s physical-address width by executing CPUID with
80000008H in EAX. The physical-address width is returned in bits 7:0 of EAX.
1. The bits corresponding to CR0.NW (bit 29) and CR0.CD (bit 30) are never checked because the
values of these bits are not changed by VM exit; see Section 24.5.1.
2. Software can determine a processor’s physical-address width by executing CPUID with
80000008H in EAX. The physical-address width is returned in bits 7:0 of EAX.
3. Bit 63 of the CR3 field in the host-state area must be 0. This is true even though, If CR4.PCIDE =
1, bit 63 of the source operand to MOV to CR3 is used to determine whether cached translation
information is invalidated.
23-8 Vol. 3B
VM ENTRIES
•
If the “load IA32_EFER” VM-exit control is 1, bits reserved in the IA32_EFER MSR
must be 0 in the field for that register. In addition, the values of the LMA and LME
bits in the field must each be that of the “host address-space size” VM-exit
control.
23.2.3
Checks on Host Segment and Descriptor-Table Registers
The following checks are performed on fields in the host-state area that correspond
to segment and descriptor-table registers:
•
In the selector field for each of CS, SS, DS, ES, FS, GS and TR, the RPL (bits 1:0)
and the TI flag (bit 2) must be 0.
•
•
The selector fields for CS and TR cannot be 0000H.
•
On processors that support Intel 64 architecture, the base-address fields for FS,
GS, GDTR, IDTR, and TR must contain canonical addresses.
The selector field for SS cannot be 0000H if the “host address-space size” VM-exit
control is 0.
23.2.4
Checks Related to Address-Space Size
On processors that support Intel 64 architecture, the following checks related to
address-space size are performed on VMX controls and fields in the host-state area:
•
If the logical processor is outside IA-32e mode (if IA32_EFER.LMA = 0) at the
time of VM entry, the following must hold:
— The “IA-32e mode guest” VM-entry control is 0.
— The “host address-space size” VM-exit control is 0.
•
If the logical processor is in IA-32e mode (if IA32_EFER.LMA = 1) at the time of
VM entry, the “host address-space size” VM-exit control must be 1.
•
If the “host address-space size” VM-exit control is 0, the following must hold:
— The “IA-32e mode guest” VM-entry control is 0.
— Bit 17 of the CR4 field (corresponding to CR4.PCIDE) is 0.
— Bits 63:32 in the RIP field is 0.
•
If the “host address-space size” VM-exit control is 1, the following must hold:
— Bit 5 of the CR4 field (corresponding to CR4.PAE) is 1.
— The RIP field contains a canonical address.
On processors that do not support Intel 64 architecture, checks are performed to
ensure that the “IA-32e mode guest” VM-entry control and the “host address-space
size” VM-exit control are both 0.
Vol. 3B 23-9
VM ENTRIES
23.3 CHECKING AND LOADING GUEST STATE
If all checks on the VMX controls and the host-state area pass (see Section 23.2), the
following operations take place concurrently: (1) the guest-state area of the VMCS is
checked to ensure that, after the VM entry completes, the state of the logical
processor is consistent with IA-32 and Intel 64 architectures; (2) processor state is
loaded from the guest-state area or as specified by the VM-entry control fields; and
(3) address-range monitoring is cleared.
Because the checking and the loading occur concurrently, a failure may be discovered only after some state has been loaded. For this reason, the logical processor
responds to such failures by loading state from the host-state area, as it would for a
VM exit. See Section 23.7.
23.3.1
Checks on the Guest State Area
This section describes checks performed on fields in the guest-state area. These
checks may be performed in any order. Some checks prevent establishment of
settings (or combinations of settings) that are currently reserved. Future processors
may allow such settings (or combinations) and may not perform the corresponding
checks. The correctness of software should not rely on VM-entry failures resulting
from the checks documented in this section.
The following subsections reference fields that correspond to processor state. Unless
otherwise stated, these references are to fields in the guest-state area.
23.3.1.1
Checks on Guest Control Registers, Debug Registers, and MSRs
The following checks are performed on fields in the guest-state area corresponding to
control registers, debug registers, and MSRs:
•
The CR0 field must not set any bit to a value not supported in VMX operation
(see Section 20.8). The following are exceptions:
— Bit 0 (corresponding to CR0.PE) and bit 31 (PG) are not checked if the
“unrestricted guest” VM-execution control is 1.1
— Bit 29 (corresponding to CR0.NW) and bit 30 (CD) are never checked
because the values of these bits are not changed by VM entry; see Section
23.3.2.1.
•
If bit 31 in the CR0 field (corresponding to PG) is 1, bit 0 in that field (PE) must
also be 1.2
•
The CR4 field must not set any bit to a value not supported in VMX operation
(see Section 20.8).
1. “Unrestricted guest” is a secondary processor-based VM-execution control. If bit 31 of the primary processor-based VM-execution controls is 0, VM entry functions as if the “unrestricted
guest” VM-execution control were 0. See Section 21.6.2.
23-10 Vol. 3B
VM ENTRIES
•
If the “load debug controls” VM-entry control is 1, bits reserved in the
IA32_DEBUGCTL MSR must be 0 in the field for that register. The first processors
to support the virtual-machine extensions supported only the 1-setting of this
control and thus performed this check unconditionally.
•
The following checks are performed on processors that support Intel 64 architecture:
— If the “IA-32e mode guest” VM-entry control is 1, bit 31 in the CR0 field
(corresponding to CR0.PG) and bit 5 in the CR4 field (corresponding to
CR4.PAE) must each be 1.1
— If the “IA-32e mode guest” VM-entry control is 0, bit 17 in the CR4 field
(corresponding to CR4.PCIDE) must each be 0.
— The CR3 field must be such that bits 63:52 and bits in the range 51:32
beyond the processor’s physical-address width are 0.2,3
— If the “load debug controls” VM-entry control is 1, bits 63:32 in the DR7 field
must be 0. The first processors to support the virtual-machine extensions
supported only the 1-setting of this control and thus performed this check
unconditionally (if they supported Intel 64 architecture).
— The IA32_SYSENTER_ESP field and the IA32_SYSENTER_EIP field must each
contain a canonical address.
•
If the “load IA32_PERF_GLOBAL_CTRL” VM-entry control is 1, bits reserved in the
IA32_PERF_GLOBAL_CTRL MSR must be 0 in the field for that register (see
Figure 30-3).
•
If the “load IA32_PAT” VM-entry control is 1, the value of the field for the
IA32_PAT MSR must be one that could be written by WRMSR without fault at CPL
0. Specifically, each of the 8 bytes in the field must have one of the values 0 (UC),
1 (WC), 4 (WT), 5 (WP), 6 (WB), or 7 (UC-).
•
If the “load IA32_EFER” VM-entry control is 1, the following checks are performed
on the field for the IA32_EFER MSR :
— Bits reserved in the IA32_EFER MSR must be 0.
2. If the capability MSR IA32_VMX_CR0_FIXED0 reports that CR0.PE must be 1 in VMX operation,
bit 0 in the CR0 field must be 1 unless the “unrestricted guest” VM-execution control and bit 31
of the primary processor-based VM-execution controls are both 1.
1. If the capability MSR IA32_VMX_CR0_FIXED0 reports that CR0.PG must be 1 in VMX operation,
bit 31 in the CR0 field must be 1 unless the “unrestricted guest” VM-execution control and bit 31
of the primary processor-based VM-execution controls are both 1.
2. Software can determine a processor’s physical-address width by executing CPUID with
80000008H in EAX. The physical-address width is returned in bits 7:0 of EAX.
3. Bit 63 of the CR3 field in the guest-state area must be 0. This is true even though, If
CR4.PCIDE = 1, bit 63 of the source operand to MOV to CR3 is used to determine whether cached
translation information is invalidated.
Vol. 3B 23-11
VM ENTRIES
— Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of the
“IA-32e mode guest” VM-exit control. It must also be identical to bit 8 (LME)
if bit 31 in the CR0 field (corresponding to CR0.PG) is 1.1
23.3.1.2
Checks on Guest Segment Registers
This section specifies the checks on the fields for CS, SS, DS, ES, FS, GS, TR, and
LDTR. The following terms are used in defining these checks:
•
The guest will be virtual-8086 if the VM flag (bit 17) is 1 in the RFLAGS field in
the guest-state area.
•
The guest will be IA-32e mode if the “IA-32e mode guest” VM-entry control is 1.
(This is possible only on processors that support Intel 64 architecture.)
•
Any one of these registers is said to be usable if the unusable bit (bit 16) is 0 in
the access-rights field for that register.
The following are the checks on these fields:
•
Selector fields.
— TR. The TI flag (bit 2) must be 0.
— LDTR. If LDTR is usable, the TI flag (bit 2) must be 0.
— SS. If the guest will not be virtual-8086 and the “unrestricted guest” VMexecution control is 0, the RPL (bits 1:0) must equal the RPL of the selector
field for CS.2
•
Base-address fields.
— CS, SS, DS, ES, FS, GS. If the guest will be virtual-8086, the address must be
the selector field shifted left 4 bits (multiplied by 16).
— The following checks are performed on processors that support Intel 64 architecture:
•
•
•
•
•
TR, FS, GS. The address must be canonical.
LDTR. If LDTR is usable, the address must be canonical.
CS. Bits 63:32 of the address must be zero.
SS, DS, ES. If the register is usable, bits 63:32 of the address must be
zero.
Limit fields for CS, SS, DS, ES, FS, GS. If the guest will be virtual-8086, the field
must be 0000FFFFH.
1. If the capability MSR IA32_VMX_CR0_FIXED0 reports that CR0.PG must be 1 in VMX operation,
bit 31 in the CR0 field must be 1 unless the “unrestricted guest” VM-execution control and bit 31
of the primary processor-based VM-execution controls are both 1.
2. “Unrestricted guest” is a secondary processor-based VM-execution control. If bit 31 of the primary processor-based VM-execution controls is 0, VM entry functions as if the “unrestricted
guest” VM-execution control were 0. See Section 21.6.2.
23-12 Vol. 3B
VM ENTRIES
•
Access-rights fields.
— CS, SS, DS, ES, FS, GS.
•
If the guest will be virtual-8086, the field must be 000000F3H. This
implies the following:
— Bits 3:0 (Type) must be 3, indicating an expand-up read/write
accessed data segment.
— Bit 4 (S) must be 1.
— Bits 6:5 (DPL) must be 3.
— Bit 7 (P) must be 1.
— Bits 11:8 (reserved), bit 12 (software available), bit 13 (reserved/L),
bit 14 (D/B), bit 15 (G), bit 16 (unusable), and bits 31:17 (reserved)
must all be 0.
•
If the guest will not be virtual-8086, the different sub-fields are
considered separately:
— Bits 3:0 (Type).
•
CS. The values allowed depend on the
“unrestricted guest” VM-execution control:
setting
of
the
— If the control is 0, the Type must be 9, 11, 13, or 15
(accessed code segment).
— If the control is 1, the Type must be either 3 (read/write
accessed expand-up data segment) or one of 9, 11, 13, and
15 (accessed code segment).
•
SS. If SS is usable, the Type must be 3 or 7 (read/write,
accessed data segment).
•
DS, ES, FS, GS. The following checks apply if the register is
usable:
— Bit 0 of the Type must be 1 (accessed).
— If bit 3 of the Type is 1 (code segment), then bit 1 of the
Type must be 1 (readable).
— Bit 4 (S). If the register is CS or if the register is usable, S must
be 1.
— Bits 6:5 (DPL).
•
CS.
— If the Type is 3 (read/write accessed expand-up data
segment), the DPL must be 0. The Type can be 3 only if the
“unrestricted guest” VM-execution control is 1.
— If the Type is 9 or 11 (non-conforming code segment), the
DPL must equal the DPL in the access-rights field for SS.
Vol. 3B 23-13
VM ENTRIES
— If the Type is 13 or 15 (conforming code segment), the DPL
cannot be greater than the DPL in the access-rights field for
SS.
•
SS.
— If the “unrestricted guest” VM-execution control is 0, the DPL
must equal the RPL from the selector field.
— The DPL must be 0 either if the Type in the access-rights field
for CS is 3 (read/write accessed expand-up data segment) or
if bit 0 in the CR0 field (corresponding to CR0.PE) is 0.1
•
DS, ES, FS, GS. The DPL cannot be less than the RPL in the
selector field if (1) the “unrestricted guest” VM-execution control
is 0; (2) the register is usable; and (3) the Type in the accessrights field is in the range 0 – 11 (data segment or nonconforming code segment).
— Bit 7 (P). If the register is CS or if the register is usable, P must be 1.
— Bits 11:8 (reserved). If the register is CS or if the register is usable,
these bits must all be 0.
— Bit 14 (D/B). For CS, D/B must be 0 if the guest will be IA-32e mode
and the L bit (bit 13) in the access-rights field is 1.
— Bit 15 (G). The following checks apply if the register is CS or if the
register is usable:
•
•
If any bit in the limit field in the range 11:0 is 0, G must be 0.
If any bit in the limit field in the range 31:20 is 1, G must be 1.
— Bits 31:17 (reserved). If the register is CS or if the register is
usable, these bits must all be 0.
— TR. The different sub-fields are considered separately:
•
Bits 3:0 (Type).
— If the guest will not be IA-32e mode, the Type must be 3 (16-bit
busy TSS) or 11 (32-bit busy TSS).
— If the guest will be IA-32e mode, the Type must be 11 (64-bit busy
TSS).
•
•
•
Bit 4 (S). S must be 0.
Bit 7 (P). P must be 1.
Bits 11:8 (reserved). These bits must all be 0.
1. The following apply if either the “unrestricted guest” VM-execution control or bit 31 of the primary processor-based VM-execution controls is 0: (1) bit 0 in the CR0 field must be 1 if the capability MSR IA32_VMX_CR0_FIXED0 reports that CR0.PE must be 1 in VMX operation; and (2) the
Type in the access-rights field for CS cannot be 3.
23-14 Vol. 3B
VM ENTRIES
•
Bit 15 (G).
— If any bit in the limit field in the range 11:0 is 0, G must be 0.
— If any bit in the limit field in the range 31:20 is 1, G must be 1.
•
•
Bit 16 (Unusable). The unusable bit must be 0.
Bits 31:17 (reserved). These bits must all be 0.
— LDTR. The following checks on the different sub-fields apply only if LDTR is
usable:
•
•
•
•
•
Bits 3:0 (Type). The Type must be 2 (LDT).
Bit 4 (S). S must be 0.
Bit 7 (P). P must be 1.
Bits 11:8 (reserved). These bits must all be 0.
Bit 15 (G).
— If any bit in the limit field in the range 11:0 is 0, G must be 0.
— If any bit in the limit field in the range 31:20 is 1, G must be 1.
•
23.3.1.3
Bits 31:17 (reserved). These bits must all be 0.
Checks on Guest Descriptor-Table Registers
The following checks are performed on the fields for GDTR and IDTR:
•
On processors that support Intel 64 architecture, the base-address fields must
contain canonical addresses.
•
Bits 31:16 of each limit field must be 0.
23.3.1.4
Checks on Guest RIP and RFLAGS
The following checks are performed on fields in the guest-state area corresponding to
RIP and RFLAGS:
•
RIP. The following checks are performed on processors that support Intel 64
architecture:
— Bits 63:32 must be 0 if the “IA-32e mode guest” VM-entry control is 0 or if
the L bit (bit 13) in the access-rights field for CS is 0.
— If the processor supports N < 64 linear-address bits, bits 63:N must be
identical if the “IA-32e mode guest” VM-entry control is 1 and the L bit in the
access-rights field for CS is 1.1 (No check applies if the processor supports 64
linear-address bits.)
1. Software can determine the number N by executing CPUID with 80000008H in EAX. The number of linear-address bits supported is returned in bits 15:8 of EAX.
Vol. 3B 23-15
VM ENTRIES
•
RFLAGS.
— Reserved bits 63:22 (bits 31:22 on processors that do not support Intel 64
architecture), bit 15, bit 5 and bit 3 must be 0 in the field, and reserved bit 1
must be 1.
— The VM flag (bit 17) must be 0 either if the “IA-32e mode guest” VM-entry
control is 1 or if bit 0 in the CR0 field (corresponding to CR0.PE) is 0.1
— The IF flag (RFLAGS[bit 9]) must be 1 if the valid bit (bit 31) in the VM-entry
interruption-information field is 1 and the interruption type (bits 10:8) is
external interrupt.
23.3.1.5
Checks on Guest Non-Register State
The following checks are performed on fields in the guest-state area corresponding to
non-register state:
•
Activity state.
— The activity-state field must contain a value in the range 0 – 3, indicating an
activity state supported by the implementation (see Section 21.4.2). Future
processors may include support for other activity states. Software should
read the VMX capability MSR IA32_VMX_MISC (see Appendix G.6) to
determine what activity states are supported.
— The activity-state field must not indicate the HLT state if the DPL (bits 6:5) in
the access-rights field for SS is not 0.2
— The activity-state field must indicate the active state if the interruptibilitystate field indicates blocking by either MOV-SS or by STI (if either bit 0 or
bit 1 in that field is 1).
— If the valid bit (bit 31) in the VM-entry interruption-information field is 1, the
interruption to be delivered (as defined by interruption type and vector) must
not be one that would normally be blocked while a logical processor is in the
activity state corresponding to the contents of the activity-state field. The
following items enumerate the interruptions (as specified in the VM-entry
interruption-information field) whose injection is allowed for the different
activity states:
•
•
Active. Any interruption is allowed.
HLT. The only events allowed are the following:
1. If the capability MSR IA32_VMX_CR0_FIXED0 reports that CR0.PE must be 1 in VMX operation,
bit 0 in the CR0 field must be 1 unless the “unrestricted guest” VM-execution control and bit 31
of the primary processor-based VM-execution controls are both 1.
2. As noted in Section 21.4.1, SS.DPL corresponds to the logical processor’s current privilege level
(CPL).
23-16 Vol. 3B
VM ENTRIES
— Those with interruption type external interrupt or non-maskable
interrupt (NMI).
— Those with interruption type hardware exception and vector 1
(debug exception) or vector 18 (machine-check exception).
— Those with interruption type other event and vector 0 (pending MTF
VM exit).
See Table 21-12 in Section 21.8.3 for details regarding the format of the
VM-entry interruption-information field.
•
•
Shutdown. Only NMIs and machine-check exceptions are allowed.
Wait-for-SIPI. No interruptions are allowed.
— The activity-state field must not indicate the wait-for-SIPI state if the “entry
to SMM” VM-entry control is 1.
•
Interruptibility state.
— The reserved bits (bits 31:4) must be 0.
— The field cannot indicate blocking by both STI and MOV SS (bits 0 and 1
cannot both be 1).
— Bit 0 (blocking by STI) must be 0 if the IF flag (bit 9) is 0 in the RFLAGS field.
— Bit 0 (blocking by STI) and bit 1 (blocking by MOV-SS) must both be 0 if the
valid bit (bit 31) in the VM-entry interruption-information field is 1 and the
interruption type (bits 10:8) in that field has value 0, indicating external
interrupt.
— Bit 1 (blocking by MOV-SS) must be 0 if the valid bit (bit 31) in the VM-entry
interruption-information field is 1 and the interruption type (bits 10:8) in that
field has value 2, indicating non-maskable interrupt (NMI).
— Bit 2 (blocking by SMI) must be 0 if the processor is not in SMM.
— Bit 2 (blocking by SMI) must be 1 if the “entry to SMM” VM-entry control is 1.
— A processor may require bit 0 (blocking by STI) to be 0 if the valid bit (bit 31)
in the VM-entry interruption-information field is 1 and the interruption type
(bits 10:8) in that field has value 2, indicating NMI. Other processors may not
make this requirement.
— Bit 3 (blocking by NMI) must be 0 if the “virtual NMIs” VM-execution control
is 1, the valid bit (bit 31) in the VM-entry interruption-information field is 1,
and the interruption type (bits 10:8) in that field has value 2 (indicating
NMI).
NOTE
If the “virtual NMIs” VM-execution control is 0, there is no
requirement that bit 3 be 0 if the valid bit in the VM-entry
interruption-information field is 1 and the interruption type in that
field has value 2.
Vol. 3B 23-17
VM ENTRIES
•
Pending debug exceptions.
— Bits 11:4, bit 13, and bits 63:15 (bits 31:15 on processors that do not
support Intel 64 architecture) must be 0.
— The following checks are performed if any of the following holds: (1) the
interruptibility-state field indicates blocking by STI (bit 0 in that field is 1);
(2) the interruptibility-state field indicates blocking by MOV SS (bit 1 in that
field is 1); or (3) the activity-state field indicates HLT:
•
•
Bit 14 (BS) must be 1 if the TF flag (bit 8) in the RFLAGS field is 1 and the
BTF flag (bit 1) in the IA32_DEBUGCTL field is 0.
•
Bit 14 (BS) must be 0 if the TF flag (bit 8) in the RFLAGS field is 0 or the
BTF flag (bit 1) in the IA32_DEBUGCTL field is 1.
VMCS link pointer. The following checks apply if the field contains a value other
than FFFFFFFF_FFFFFFFFH:
— Bits 11:0 must be 0.
— Bits beyond the processor’s physical-address width must be 0.1,2
— The 32 bits located in memory referenced by the value of the field (as a
physical address) must contain the processor’s VMCS revision identifier (see
Section 21.2).
— If the processor is not in SMM or the “entry to SMM” VM-entry control is 1, the
field must not contain the current VMCS pointer.
— If the processor is in SMM and the “entry to SMM” VM-entry control is 0, the
field must not contain the VMXON pointer.
23.3.1.6
Checks on Guest Page-Directory-Pointer-Table Entries
If CR0.PG =1, CR4.PAE = 1, and IA32_EFER.LMA = 0, the logical processor also uses
PAE paging (see Section 4.4 in the Intel® 64 and IA-32 Architectures Software
Developer’s Manual, Volume 3A).3 When PAE paging is in use, the physical address in
CR3 references a table of page-directory-pointer-table entries (PDPTEs). A MOV
to CR3 when PAE paging is in use checks the validity of the PDPTEs.
A VM entry is to a guest that uses PAE paging if (1) bit 31 (corresponding to CR0.PG)
is set in the CR0 field in the guest-state area; (2) bit 5 (corresponding to CR4.PAE) is
1. Software can determine a processor’s physical-address width by executing CPUID with
80000008H in EAX. The physical-address width is returned in bits 7:0 of EAX.
2. If IA32_VMX_BASIC[48] is read as 1, this field must not set any bits in the range 63:32; see
Appendix G.1.
3. On processors that support Intel 64 architecture, the physical-address extension may support
more than 36 physical-address bits. Software can determine the number physical-address bits
supported by executing CPUID with 80000008H in EAX. The physical-address width is returned
in bits 7:0 of EAX.
23-18 Vol. 3B
VM ENTRIES
set in the CR4 field; and (3) the “IA-32e mode guest” VM-entry control is 0. Such a
VM entry checks the validity of the PDPTEs:
•
If the “enable EPT” VM-execution control is 0, VM entry checks the validity of the
PDPTEs referenced by the CR3 field in the guest-state area if either (1) PAE
paging was not in use before the VM entry; or (2) the value of CR3 is changing as
a result of the VM entry. VM entry may check their validity even if neither (1) nor
(2) hold.1
•
If the “enable EPT” VM-execution control is 1, VM entry checks the validity of the
PDPTE fields in the guest-state area (see Section 21.4.2).
A VM entry to a guest that does not use PAE paging does not check the validity of any
PDPTEs.
A VM entry that checks the validity of the PDPTEs uses the same checks that are used
when CR3 is loaded with MOV to CR3 when PAE paging is in use.2 If MOV to CR3
would cause a general-protection exception due to the PDPTEs that would be loaded
(e.g., because a reserved bit is set), the VM entry fails.
23.3.2 Loading Guest State
Processor state is updated on VM entries in the following ways:
•
•
•
Some state is loaded from the guest-state area.
Some state is determined by VM-entry controls.
The page-directory pointers are loaded based on the values of certain control
registers.
This loading may be performed in any order and in parallel with the checking of VMCS
contents (see Section 23.3.1).
The loading of guest state is detailed in Section 23.3.2.1 to Section 23.3.2.4. These
sections reference VMCS fields that correspond to processor state. Unless otherwise
stated, these references are to fields in the guest-state area.
In addition to the state loading described in this section, VM entries may load MSRs
from the VM-entry MSR-load area (see Section 23.4). This loading occurs only after
the state loading described in this section and the checking of VMCS contents
described in Section 23.3.1.
1. “Enable EPT” is a secondary processor-based VM-execution control. If bit 31 of the primary processor-based VM-execution controls is 0, VM entry functions as if the “enable EPT” VM-execution control were 0. See Section 21.6.2.
2. This implies that (1) bits 11:9 in each PDPTE are ignored; and (2) if bit 0 (present) is clear in one
of the PDPTEs, bits 63:1 of that PDPTE are ignored.
Vol. 3B 23-19
VM ENTRIES
23.3.2.1
Loading Guest Control Registers, Debug Registers, and MSRs
The following items describe how guest control registers, debug registers, and MSRs
are loaded on VM entry:
•
CR0 is loaded from the CR0 field with the exception of the following bits, which
are never modified on VM entry: ET (bit 4); reserved bits 15:6, 17, and 28:19;
NW (bit 29) and CD (bit 30).1 The values of these bits in the CR0 field are
ignored.
•
•
CR3 and CR4 are loaded from the CR3 field and the CR4 field, respectively.
If the “load debug controls” VM-execution control is 1, DR7 is loaded from the
DR7 field with the exception that bit 12 and bits 15:14 are always 0 and bit 10 is
always 1. The values of these bits in the DR7 field are ignored.
The first processors to support the virtual-machine extensions supported only
the 1-setting of the “load debug controls” VM-execution control and thus always
loaded DR7 from the DR7 field.
•
The following describes how some MSRs are loaded using fields in the guest-state
area:
— If the “load debug controls” VM-execution control is 1, the IA32_DEBUGCTL
MSR is loaded from the IA32_DEBUGCTL field. The first processors to support
the virtual-machine extensions supported only the 1-setting of this control
and thus always loaded the IA32_DEBUGCTL MSR from the IA32_DEBUGCTL
field.
— The IA32_SYSENTER_CS MSR is loaded from the IA32_SYSENTER_CS field.
Since this field has only 32 bits, bits 63:32 of the MSR are cleared to 0.
— The IA32_SYSENTER_ESP and IA32_SYSENTER_EIP MSRs are loaded from
the IA32_SYSENTER_ESP field and the IA32_SYSENTER_EIP field, respectively. On processors that do not support Intel 64 architecture, these fields
have only 32 bits; bits 63:32 of the MSRs are cleared to 0.
— The following are performed on processors that support Intel 64 architecture:
•
The MSRs FS.base and GS.base are loaded from the base-address fields
for FS and GS, respectively (see Section 23.3.2.2).
•
If the “load IA32_EFER” VM-entry control is 0, bits in the IA32_EFER MSR
are modified as follows:
— IA32_EFER.LMA is loaded with the setting of the “IA-32e mode
guest” VM-entry control.
— If CR0 is being loaded so that CR0.PG = 1, IA32_EFER.LME is also
loaded with the setting of the “IA-32e mode guest” VM-entry
control.2 Otherwise, IA32_EFER.LME is unmodified.
1. Bits 15:6, bit 17, and bit 28:19 of CR0 and CR0.ET are unchanged by executions of MOV to CR0.
Bits 15:6, bit 17, and bit 28:19 of CR0 are always 0 and CR0.ET is always 1.
23-20 Vol. 3B
VM ENTRIES
See below for the case in which the “load IA32_EFER” VM-entry control is
1
— If the “load IA32_PERF_GLOBAL_CTRL” VM-entry control is 1, the
IA32_PERF_GLOBAL_CTRL MSR is loaded from the
IA32_PERF_GLOBAL_CTRL field.
— If the “load IA32_PAT” VM-entry control is 1, the IA32_PAT MSR is loaded
from the IA32_PAT field.
— If the “load IA32_EFER” VM-entry control is 1, the IA32_EFER MSR is loaded
from the IA32_EFER field.
With the exception of FS.base and GS.base, any of these MSRs is subsequently
overwritten if it appears in the VM-entry MSR-load area. See Section 23.4.
•
The SMBASE register is unmodified by all VM entries except those that return
from SMM.
23.3.2.2
Loading Guest Segment Registers and Descriptor-Table Registers
For each of CS, SS, DS, ES, FS, GS, TR, and LDTR, fields are loaded from the gueststate area as follows:
•
The unusable bit is loaded from the access-rights field. This bit can never be set
for TR (see Section 23.3.1.2). If it is set for one of the other registers, the
following apply:
— For each of CS, SS, DS, ES, FS, and GS, uses of the segment cause faults
(general-protection exception or stack-fault exception) outside 64-bit mode,
just as they would had the segment been loaded using a null selector. This bit
does not cause accesses to fault in 64-bit mode.
— If this bit is set for LDTR, uses of LDTR cause general-protection exceptions in
all modes, just as they would had LDTR been loaded using a null selector.
If this bit is clear for any of CS, SS, DS, ES, FS, GS, TR, and LDTR, a null
selector value does not cause a fault (general-protection exception or stackfault exception).
•
•
TR. The selector, base, limit, and access-rights fields are loaded.
CS.
— The following fields are always loaded: selector, base address, limit, and
(from the access-rights field) the L, D, and G bits.
— For the other fields, the unusable bit of the access-rights field is consulted:
•
If the unusable bit is 0, all of the access-rights field is loaded.
2. If the capability MSR IA32_VMX_CR0_FIXED0 reports that CR0.PG must be 1 in VMX operation,
VM entry must be loading CR0 so that CR0.PG = 1 unless the “unrestricted guest” VM-execution
control and bit 31 of the primary processor-based VM-execution controls are both 1.
Vol. 3B 23-21
VM ENTRIES
•
•
If the unusable bit is 1, the remainder of CS access rights are undefined
after VM entry.
SS, DS, ES, FS, GS, and LDTR.
— The selector fields are loaded.
— For the other fields, the unusable bit of the corresponding access-rights field
is consulted:
•
If the unusable bit is 0, the base-address, limit, and access-rights fields
are loaded.
•
If the unusable bit is 1, the base address, the segment limit, and the
remainder of the access rights are undefined after VM entry with the
following exceptions:
— Bits 3:0 of the base address for SS are cleared to 0.
— SS.DPL is always loaded from the SS access-rights field. This will be
the current privilege level (CPL) after the VM entry completes.
— SS.B is always set to 1.
— The base addresses for FS and GS are loaded from the corresponding fields in the VMCS. On processors that support Intel 64
architecture, the values loaded for base addresses for FS and GS are
also manifest in the FS.base and GS.base MSRs.
— On processors that support Intel 64 architecture, the base address
for LDTR is set to an undefined but canonical value.
— On processors that support Intel 64 architecture, bits 63:32 of the
base addresses for SS, DS, and ES are cleared to 0.
GDTR and IDTR are loaded using the base and limit fields.
23.3.2.3
Loading Guest RIP, RSP, and RFLAGS
RSP, RIP, and RFLAGS are loaded from the RSP field, the RIP field, and the RFLAGS
field, respectively. The following items regard the upper 32 bits of these fields on
VM entries that are not to 64-bit mode:
•
Bits 63:32 of RSP are undefined outside 64-bit mode. Thus, a logical processor
may ignore the contents of bits 63:32 of the RSP field on VM entries that are not
to 64-bit mode.
•
As noted in Section 23.3.1.4, bits 63:32 of the RIP and RFLAGS fields must be 0
on VM entries that are not to 64-bit mode.
23.3.2.4
Loading Page-Directory-Pointer-Table Entries
As noted in Section 23.3.1.6, the logical processor uses PAE paging if bit 5 in CR4
(CR4.PAE) is 1 and IA32_EFER.LMA is 0. A VM entry to a guest that uses PAE paging
23-22 Vol. 3B
VM ENTRIES
loads the PDPTEs into internal, non-architectural registers based on the setting of the
“enable EPT” VM-execution control:
•
If the control is 0, the PDPTEs are loaded from the page-directory-pointer table
referenced by the physical address in the value of CR3 being loaded by the
VM entry (see Section 23.3.2.1). The values loaded are treated as physical
addresses in VMX non-root operation.
•
If the control is 1, the PDPTEs are loaded from corresponding fields in the gueststate area (see Section 21.4.2). The values loaded are treated as guest-physical
addresses in VMX non-root operation.
23.3.2.5
Updating Non-Register State
Section 25.3 describe how the VMX architecture controls how a logical processor
manages information in the TLBs and paging-structure caches. The following items
detail how VM entries invalidate cached mappings:
•
If the “enable VPID” VM-execution control is 0, the logical processor invalidates
linear mappings and combined mappings associated with VPID 0000H (for all
PCIDs); combined mappings for VPID 0000H are invalidated for all EP4TA values
(EP4TA is the value of bits 51:12 of EPTP).
•
VM entries are not required to invalidate any guest-physical mappings, nor are
they required to invalidate any linear mappings or combined mappings if the
“enable VPID” VM-execution control is 1.
23.3.3
Clearing Address-Range Monitoring
The Intel 64 and IA-32 architectures allow software to monitor a specified address
range using the MONITOR and MWAIT instructions. See Section 8.10.4 in the Intel®
64 and IA-32 Architectures Software Developer’s Manual, Volume 3A. VM entries
clear any address-range monitoring that may be in effect.
23.4
LOADING MSRS
VM entries may load MSRs from the VM-entry MSR-load area (see Section 21.8.2).
Specifically each entry in that area (up to the number specified in the VM-entry MSRload count) is processed in order by loading the MSR indexed by bits 31:0 with the
contents of bits 127:64 as they would be written by WRMSR.1
Processing of an entry fails in any of the following cases:
•
The value of bits 31:0 is either C0000100H (the IA32_FS_BASE MSR) or
C0000101 (the IA32_GS_BASE MSR).
1. Because attempts to modify the value of IA32_EFER.LMA by WRMSR are ignored, attempts to
modify it using the VM-entry MSR-load area are also ignored.
Vol. 3B 23-23
VM ENTRIES
•
The value of bits 31:8 is 000008H, meaning that the indexed MSR is one that
allows access to an APIC register when the local APIC is in x2APIC mode.
•
The value of bits 31:0 indicates an MSR that can be written only in systemmanagement mode (SMM) and the VM entry did not commence in SMM.
(IA32_SMM_MONITOR_CTL is an MSR that can be written only in SMM.)
•
The value of bits 31:0 indicates an MSR that cannot be loaded on VM entries for
model-specific reasons. A processor may prevent loading of certain MSRs even if
they can normally be written by WRMSR. Such model-specific behavior is
documented in Appendix B.
•
•
Bits 63:32 are not all 0.
An attempt to write bits 127:64 to the MSR indexed by bits 31:0 of the entry
would cause a general-protection exception if executed via WRMSR with
CPL = 0.1
The VM entry fails if processing fails for any entry. The logical processor responds to
such failures by loading state from the host-state area, as it would for a VM exit. See
Section 23.7.
If any MSR is being loaded in such a way that would architecturally require a TLB
flush, the TLBs are updated so that, after VM entry, the logical processor will not use
any translations that were cached before the transition.
23.5
EVENT INJECTION
If the valid bit in the VM-entry interruption-information field (see Section 21.8.3) is
1, VM entry causes an event to be delivered (or made pending) after all components
of guest state have been loaded (including MSRs) and after the VM-execution control
fields have been established.
•
If the interruption type in the field is 0 (external interrupt), 2 (non-maskable
interrupt); 3 (hardware exception), 4 (software interrupt), 5 (privileged software
exception), or 6 (software exception), the event is delivered as described in
Section 23.5.1.
•
If the interruption type in the field is 7 (other event) and the vector field is 0, an
MTF VM exit is pending after VM entry. See Section 23.5.2.
23.5.1
Vectored-Event Injection
VM entry delivers an injected vectored event within the guest context established by
VM entry. This means that delivery occurs after all components of guest state have
1. If CR0.PG = 1, WRMSR to the IA32_EFER MSR causes a general-protection exception if it would
modify the LME bit. If VM entry has established CR0.PG = 1, the IA32_EFER MSR should not be
included in the VM-entry MSR-load area for the purpose of modifying the LME bit.
23-24 Vol. 3B
VM ENTRIES
been loaded (including MSRs) and after the VM-execution control fields have been
established.1 The event is delivered using the vector in that field to select a
descriptor in the IDT. Since event injection occurs after loading IDTR from the gueststate area, this is the guest IDT.
Section 23.5.1.1 provides details of vectored-event injection. In general, the event is
delivered exactly as if it had been generated normally.
If event delivery encounters a nested exception (for example, a general-protection
exception because the vector indicates a descriptor beyond the IDT limit), the exception bitmap is consulted using the vector of that exception. If the bit is 0, the exception is delivered through the IDT. If the bit is 1, a VM exit occurs. Section 23.5.1.2
details cases in which event injection causes a VM exit.
23.5.1.1
Details of Vectored-Event Injection
The event-injection process is controlled by the contents of the VM-entry interruption
information field (format given in Table 21-12), the VM-entry exception error-code
field, and the VM-entry instruction-length field. The following items provide details of
the process:
•
The value pushed on the stack for RFLAGS is generally that which was loaded
from the guest-state area. The value pushed for the RF flag is not modified based
on the type of event being delivered. However, the pushed value of RFLAGS may
be modified if a software interrupt is being injected into a guest that will be in
virtual-8086 mode (see below). After RFLAGS is pushed on the stack, the value
in the RFLAGS register is modified as is done normally when delivering an event
through the IDT.
•
The instruction pointer that is pushed on the stack depends on the type of event
and whether nested exceptions occur during its delivery. The term current
guest RIP refers to the value to be loaded from the guest-state area. The value
pushed is determined as follows:2
— If VM entry successfully injects (with no nested exception) an event with
interruption type external interrupt, NMI, or hardware exception, the current
guest RIP is pushed on the stack.
— If VM entry successfully injects (with no nested exception) an event with
interruption type software interrupt, privileged software exception, or
software exception, the current guest RIP is incremented by the VM-entry
instruction length before being pushed on the stack.
1. This does not imply that injection of an exception or interrupt will cause a VM exit due to the settings of VM-execution control fields (such as the exception bitmap) that would cause a VM exit if
the event had occurred in VMX non-root operation. In contrast, a nested exception encountered
during event delivery may cause a VM exit; see Section 23.5.1.1.
2. While these items refer to RIP, the width of the value pushed (16 bits, 32 bits, or 64 bits) is
determined normally.
Vol. 3B 23-25
VM ENTRIES
— If VM entry encounters an exception while injecting an event and that
exception does not cause a VM exit, the current guest RIP is pushed on the
stack regardless of event type or VM-entry instruction length. If the
encountered exception does cause a VM exit that saves RIP, the saved RIP is
current guest RIP.
•
If the deliver-error-code bit (bit 11) is set in the VM-entry interruptioninformation field, the contents of the VM-entry exception error-code field is
pushed on the stack as an error code would be pushed during delivery of an
exception.
•
DR6, DR7, and the IA32_DEBUGCTL MSR are not modified by event injection,
even if the event has vector 1 (normal deliveries of debug exceptions, which have
vector 1, do update these registers).
•
If VM entry is injecting a software interrupt and the guest will be in virtual-8086
mode (RFLAGS.VM = 1), no general-protection exception can occur due to
RFLAGS.IOPL < 3. A VM monitor should check RFLAGS.IOPL before injecting
such an event and, if desired, inject a general-protection exception instead of a
software interrupt.
•
If VM entry is injecting a software interrupt and the guest will be in virtual-8086
mode with virtual-8086 mode extensions (RFLAGS.VM = CR4.VME = 1), event
delivery is subject to VME-based interrupt redirection based on the software
interrupt redirection bitmap in the task-state segment (TSS) as follows:
— If bit n in the bitmap is clear (where n is the number of the software
interrupt), the interrupt is directed to an 8086 program interrupt handler: the
processor uses a 16-bit interrupt-vector table (IVT) located at linear address
zero. If the value of RFLAGS.IOPL is less than 3, the following modifications
are made to the value of RFLAGS that is pushed on the stack: IOPL is set to
3, and IF is set to the value of VIF.
— If bit n in the bitmap is set (where n is the number of the software interrupt),
the interrupt is directed to a protected-mode interrupt handler. (In other
words, the injection is treated as described in the next item.) In this case, the
software interrupt does not invoke such a handler if RFLAGS.IOPL < 3 (a
general-protection exception occurs instead). However, as noted above,
RFLAGS.IOPL cannot cause an injected software interrupt to cause such a
exception. Thus, in this case, the injection invokes a protected-mode
interrupt handler independent of the value of RFLAGS.IOPL.
Injection of events of other types are not subject to this redirection.
•
If VM entry is injecting a software interrupt (not redirected as described above)
or software exception, privilege checking is performed on the IDT descriptor
being accessed as would be the case for executions of INT n, INT3, or INTO (the
descriptor’s DPL cannot be less than CPL). There is no checking of RFLAGS.IOPL,
even if the guest will be in virtual-8086 mode. Failure of this check may lead to a
nested exception. Injection of an event with interruption type external interrupt,
NMI, hardware exception, and privileged software exception, or with interruption
23-26 Vol. 3B
VM ENTRIES
type software interrupt and being redirected as described above, do not perform
these checks.
•
If VM entry is injecting a non-maskable interrupt (NMI) and the “virtual NMIs”
VM-execution control is 1, virtual-NMI blocking is in effect after VM entry.
•
The transition causes a last-branch record to be logged if the LBR bit is set in the
IA32_DEBUGCTL MSR. This is true even for events such as debug exceptions,
which normally clear the LBR bit before delivery.
•
The last-exception record MSRs (LERs) may be updated based on the setting of
the LBR bit in the IA32_DEBUGCTL MSR. Events such as debug exceptions, which
normally clear the LBR bit before they are delivered, and therefore do not
normally update the LERs, may do so as part of VM-entry event injection.
•
If injection of an event encounters a nested exception that does not itself cause a
VM exit, the value of the EXT bit (bit 0) in any error code pushed on the stack is
determined as follows:
— If event being injected has interruption type external interrupt, NMI,
hardware exception, or privileged software exception and encounters a
nested exception (but does not produce a double fault), the error code for the
first such exception encountered sets the EXT bit.
— If event being injected is a software interrupt or an software exception and
encounters a nested exception (but does not produce a double fault), the
error code for the first such exception encountered clears the EXT bit.
— If event delivery encounters a nested exception and delivery of that
exception encounters another exception (but does not produce a double
fault), the error code for that exception sets the EXT bit. If a double fault is
produced, the error code for the double fault is 0000H (the EXT bit is clear).
23.5.1.2
VM Exits During Event Injection
An event being injected never causes a VM exit directly regardless of the settings of
the VM-execution controls. For example, setting the “NMI exiting” VM-execution
control to 1 does not cause a VM exit due to injection of an NMI.
However, the event-delivery process may lead to a VM exit:
•
If the vector in the VM-entry interruption-information field identifies a task gate
in the IDT, the attempted task switch may cause a VM exit just as it would had
the injected event occurred during normal execution in VMX non-root operation
(see Section 22.6.2).
•
If event delivery encounters a nested exception, a VM exit may occur depending
on the contents of the exception bitmap (see Section 22.3).
•
If event delivery generates a double-fault exception (due to a nested exception);
the logical processor encounters another nested exception while attempting to
call the double-fault handler; and that exception does not cause a VM exit due to
the exception bitmap; then a VM exit occurs due to triple fault (see Section
22.3).
Vol. 3B 23-27
VM ENTRIES
•
If event delivery injects a double-fault exception and encounters a nested
exception that does not cause a VM exit due to the exception bitmap, then a
VM exit occurs due to triple fault (see Section 22.3).
•
If the “virtualize APIC accesses” VM-execution control is 1 and event delivery
generates an access to the APIC-access page, that access may cause an APICaccess VM exit (see Section 22.2) or, if the access is a VTPR access, be treated as
specified in Section 22.5.3.1
If the event-delivery process does cause a VM exit, the processor state before the
VM exit is determined just as it would be had the injected event occurred during
normal execution in VMX non-root operation. If the injected event directly accesses a
task gate that cause a VM exit or if the first nested exception encountered causes a
VM exit, information about the injected event is saved in the IDT-vectoring information field (see Section 24.2.3).
23.5.1.3
Event Injection for VM Entries to Real-Address Mode
If VM entry is loading CR0.PE with 0, any injected vectored event is delivered as
would normally be done in real-address mode.2 Specifically, VM entry uses the vector
provided in the VM-entry interruption-information field to select a 4-byte entry from
an interrupt-vector table at the linear address in IDTR.base. Further details are
provided in Section 15.1.4 in Volume 3A of the IA-32 Intel® Architecture Software
Developer’s Manual.
Because bit 11 (deliver error code) in the VM-entry interruption-information field
must be 0 if CR0.PE will be 0 after VM entry (see Section 23.2.1.3), vectored events
injected with CR0.PE = 0 do not push an error code on the stack. This is consistent
with event delivery in real-address mode.
If event delivery encounters a fault (due to a violation of IDTR.limit or of SS.limit),
the fault is treated as if it had occurred during event delivery in VMX non-root operation. Such a fault may lead to a VM exit as discussed in Section 23.5.1.2.
23.5.2
Injection of Pending MTF VM Exits
If the interruption type in the VM-entry interruption-information field is 7 (other
event) and the vector field is 0, VM entry causes an MTF VM exit to be pending on the
instruction boundary following VM entry. This is the case even if the “monitor trap
flag” VM-execution control is 0. See Section 22.7.2 for the treatment of pending MTF
VM exits.
1. “Virtualize APIC accesses” is a secondary processor-based VM-execution control. If bit 31 of the
primary processor-based VM-execution controls is 0, VM entry functions as if the “virtualize APIC
accesses” VM-execution control were 0. See Section 21.6.2.
2. If the capability MSR IA32_VMX_CR0_FIXED0 reports that CR0.PE must be 1 in VMX operation,
VM entry must be loading CR0.PE with 1 unless the “unrestricted guest” VM-execution control
and bit 31 of the primary processor-based VM-execution controls are both 1.
23-28 Vol. 3B
VM ENTRIES
23.6
SPECIAL FEATURES OF VM ENTRY
This section details a variety of features of VM entry. It uses the following terminology: a VM entry is vectoring if the valid bit (bit 31) of the VM-entry interruption
information field is 1 and the interruption type in the field is 0 (external interrupt), 2
(non-maskable interrupt); 3 (hardware exception), 4 (software interrupt), 5 (privileged software exception), or 6 (software exception).
23.6.1
Interruptibility State
The interruptibility-state field in the guest-state area (see Table 21-3) contains bits
that control blocking by STI, blocking by MOV SS, and blocking by NMI. This field
impacts event blocking after VM entry as follows:
•
If the VM entry is vectoring, there is no blocking by STI or by MOV SS following
the VM entry, regardless of the contents of the interruptibility-state field.
•
If the VM entry is not vectoring, the following apply:
— Events are blocked by STI if and only if bit 0 in the interruptibility-state field
is 1. This blocking is cleared after the guest executes one instruction or incurs
an exception (including a debug exception made pending by VM entry; see
Section 23.6.3).
— Events are blocked by MOV SS if and only if bit 1 in the interruptibility-state
field is 1. This may affect the treatment of pending debug exceptions; see
Section 23.6.3. This blocking is cleared after the guest executes one
instruction or incurs an exception (including a debug exception made pending
by VM entry).
•
The blocking of non-maskable interrupts (NMIs) is determined as follows:
— If the “virtual NMIs” VM-execution control is 0, NMIs are blocked if and only if
bit 3 (blocking by NMI) in the interruptibility-state field is 1. If the “NMI
exiting” VM-execution control is 0, execution of the IRET instruction removes
this blocking (even if the instruction generates a fault). If the “NMI exiting”
control is 1, IRET does not affect this blocking.
— The following items describe the use of bit 3 (blocking by NMI) in the interruptibility-state field if the “virtual NMIs” VM-execution control is 1:
•
The bit’s value does not affect the blocking of NMIs after VM entry. NMIs
are not blocked in VMX non-root operation (except for ordinary blocking
for other reasons, such as by the MOV SS instruction, the wait-for-SIPI
state, etc.)
•
The bit’s value determines whether there is virtual-NMI blocking after
VM entry. If the bit is 1, virtual-NMI blocking is in effect after VM entry. If
the bit is 0, there is no virtual-NMI blocking after VM entry unless the
VM entry is injecting an NMI (see Section 23.5.1.1). Execution of IRET
removes virtual-NMI blocking (even if the instruction generates a fault).
Vol. 3B 23-29
VM ENTRIES
If the “NMI exiting” VM-execution control is 0, the “virtual NMIs” control must
be 0; see Section 23.2.1.1.
•
Blocking of system-management interrupts (SMIs) is determined as follows:
— If the VM entry was not executed in system-management mode (SMM), SMI
blocking is unchanged by VM entry.
— If the VM entry was executed in SMM, SMIs are blocked after VM entry if and
only if the bit 2 in the interruptibility-state field is 1.
23.6.2
Activity State
The activity-state field in the guest-state area controls whether, after VM entry, the
logical processor is active or in one of the inactive states identified in Section 21.4.2.
The use of this field is determined as follows:
•
If the VM entry is vectoring, the logical processor is in the active state after
VM entry. While the consistency checks described in Section 23.3.1.5 on the
activity-state field do apply in this case, the contents of the activity-state field do
not determine the activity state after VM entry.
•
If the VM entry is not vectoring, the logical processor ends VM entry in the
activity state specified in the guest-state area. If VM entry ends with the logical
processor in an inactive activity state, the VM entry generates any special bus
cycle that is normally generated when that activity state is entered from the
active state. If VM entry would end with the logical processor in the shutdown
state and the logical processor is in SMX operation,1 an Intel® TXT shutdown
condition occurs. The error code used is 0000H, indicating “legacy shutdown.”
See Intel® Trusted Execution Technology Preliminary Architecture Specification.
•
Some activity states unconditionally block certain events. The following blocking
is in effect after any VM entry that puts the processor in the indicated state:
— The active state blocks start-up IPIs (SIPIs). SIPIs that arrive while a logical
processor is in the active state and in VMX non-root operation are discarded
and do not cause VM exits.
— The HLT state blocks start-up IPIs (SIPIs). SIPIs that arrive while a logical
processor is in the HLT state and in VMX non-root operation are discarded and
do not cause VM exits.
— The shutdown state blocks external interrupts and SIPIs. External interrupts
that arrive while a logical processor is in the shutdown state and in VMX nonroot operation do not cause VM exits even if the “external-interrupt exiting”
VM-execution control is 1. SIPIs that arrive while a logical processor is in the
1. A logical processor is in SMX operation if GETSEC[SEXIT] has not been executed since the last
execution of GETSEC[SENTER]. See Chapter 6, “Safer Mode Extensions Reference,” in Intel® 64
and IA-32 Architectures Software Developer’s Manual, Volume 2B.
23-30 Vol. 3B
VM ENTRIES
shutdown state and in VMX non-root operation are discarded and do not
cause VM exits.
— The wait-for-SIPI state blocks external interrupts, non-maskable interrupts
(NMIs), INIT signals, and system-management interrupts (SMIs). Such
events do not cause VM exits if they arrive while a logical processor is in the
wait-for-SIPI state and in VMX non-root operation do not cause VM exits
regardless of the settings of the pin-based VM-execution controls.
23.6.3
Delivery of Pending Debug Exceptions after VM Entry
The pending debug exceptions field in the guest-state area indicates whether there
are debug exceptions that have not yet been delivered (see Section 21.4.2). This
section describes how these are treated on VM entry.
There are no pending debug exceptions after VM entry if any of the following are
true:
•
The VM entry is vectoring with one of the following interruption types: external
interrupt, non-maskable interrupt (NMI), hardware exception, or privileged
software exception.
•
The interruptibility-state field does not indicate blocking by MOV SS and the
VM entry is vectoring with either of the following interruption type: software
interrupt or software exception.
•
The VM entry is not vectoring and the activity-state field indicates either
shutdown or wait-for-SIPI.
If none of the above hold, the pending debug exceptions field specifies the debug
exceptions that are pending for the guest. There are valid pending debug exceptions if either the BS bit (bit 14) or the enable-breakpoint bit (bit 12) is 1. If there
are valid pending debug exceptions, they are handled as follows:
•
If the VM entry is not vectoring, the pending debug exceptions are treated as
they would had they been encountered normally in guest execution:
— If the logical processor is not blocking such exceptions (the interruptibilitystate field indicates no blocking by MOV SS), a debug exception is delivered
after VM entry (see below).
— If the logical processor is blocking such exceptions (due to blocking by
MOV SS), the pending debug exceptions are held pending or lost as would
normally be the case.
•
If the VM entry is vectoring (with interruption type software interrupt or software
exception and with blocking by MOV SS), the following items apply:
— For injection of a software interrupt or of a software exception with vector 3
(#BP) or vector 4 (#OF), the pending debug exceptions are treated as they
would had they been encountered normally in guest execution if the corresponding instruction (INT3 or INTO) were executed after a MOV SS that
encountered a debug trap.
Vol. 3B 23-31
VM ENTRIES
— For injection of a software exception with a vector other than 3 and 4, the
pending debug exceptions may be lost or they may be delivered after
injection (see below).
If there are no valid pending debug exceptions (as defined above), no pending debug
exceptions are delivered after VM entry.
If a pending debug exception is delivered after VM entry, it has the priority of “traps
on the previous instruction” (see Section 6.9 in the Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 3A). Thus, INIT signals and systemmanagement interrupts (SMIs) take priority of such an exception, as do VM exits
induced by the TPR shadow (see Section 23.6.7) and pending MTF VM exits (see
Section 23.6.8. The exception takes priority over any pending non-maskable interrupt (NMI) or external interrupt and also over VM exits due to the 1-settings of the
“interrupt-window exiting” and “NMI-window exiting” VM-execution controls.
A pending debug exception delivered after VM entry causes a VM exit if the bit 1
(#DB) is 1 in the exception bitmap. If it does not cause a VM exit, it updates DR6
normally.
23.6.4
VMX-Preemption Timer
If the “activate VMX-preemption timer” VM-execution control is 1, VM entry starts
the VMX-preemption timer with the unsigned value in the VMX-preemption timervalue field.
It is possible for the VMX-preemption timer to expire during VM entry (e.g., if the
value in the VMX-preemption timer-value field is zero). If this happens (and if the VM
entry was not to the wait-for-SIPI state), a VM exit occurs with its normal priority
after any event injection and before execution of any instruction following VM entry.
For example, any pending debug exceptions established by VM entry (see Section
23.6.3) take priority over a timer-induced VM exit. (The timer-induced VM exit will
occur after delivery of the debug exception, unless that exception or its delivery
causes a different VM exit.)
See Section 22.7.1 for details of the operation of the VMX-preemption timer in VMX
non-root operation, including the blocking and priority of the VM exits that it causes.
23.6.5
Interrupt-Window Exiting
The “interrupt-window exiting” VM-execution control may cause a VM exit to occur
immediately after VM entry (see Section 22.3 for details).
The following items detail the treatment of these VM exits:
•
•
These VM exits follow event injection if such injection is specified for VM entry.
Non-maskable interrupts (NMIs) and higher priority events take priority over
VM exits caused by this control. VM exits caused by this control take priority over
external interrupts and lower priority events.
23-32 Vol. 3B
VM ENTRIES
•
VM exits caused by this control wake the logical processor if it just entered the
HLT state because of a VM entry (see Section 23.6.2). They do not occur if the
logical processor just entered the shutdown state or the wait-for-SIPI state.
23.6.6
NMI-Window Exiting
The “NMI-window exiting” VM-execution control may cause a VM exit to occur immediately after VM entry (see Section 22.3 for details).
The following items detail the treatment of these VM exits:
•
•
These VM exits follow event injection if such injection is specified for VM entry.
•
VM exits caused by this control wake the logical processor if it just entered either
the HLT state or the shutdown state because of a VM entry (see Section 23.6.2).
They do not occur if the logical processor just entered the wait-for-SIPI state.
Debug-trap exceptions (see Section 23.6.3) and higher priority events take
priority over VM exits caused by this control. VM exits caused by this control take
priority over non-maskable interrupts (NMIs) and lower priority events.
23.6.7
VM Exits Induced by the TPR Shadow
If the “use TPR shadow” and “virtualize APIC accesses” VM-execution controls are
both 1, a VM exit occurs immediately after VM entry if the value of bits 3:0 of the TPR
threshold VM-execution control field is greater than the value of bits 7:4 in byte 80H
on the virtual-APIC page (see Section 21.6.8).1
The following items detail the treatment of these VM exits:
•
The VM exits are not blocked if RFLAGS.IF = 0 or by the setting of bits in the
interruptibility-state field in guest-state area.
•
•
The VM exits follow event injection if such injection is specified for VM entry.
•
These VM exits wake the logical processor if it just entered the HLT state as part
of a VM entry (see Section 23.6.2). They do not occur if the logical processor just
entered the shutdown state or the wait-for-SIPI state.
VM exits caused by this control take priority over system-management interrupts
(SMIs), INIT signals, and lower priority events. They thus have priority over the
VM exits described in Section 23.6.5, Section 23.6.6, and Section 23.6.8, as well
as any interrupts or debug exceptions that may be pending at the time of
VM entry.
If such a VM exit is suppressed because the processor just entered the
shutdown state, it occurs after the delivery of any event that cause the logical
1. “Virtualize APIC accesses” is a secondary processor-based VM-execution control. If bit 31 of the
primary processor-based VM-execution controls is 0, VM entry functions as if the “virtualize APIC
accesses” VM-execution control were 0. See Section 21.6.2.
Vol. 3B 23-33
VM ENTRIES
processor to leave the shutdown state while remaining in VMX non-root
operation (e.g., due to an NMI that occurs while the “NMI-exiting” VM-execution
control is 0).
•
The basic exit reason is “TPR below threshold.”
23.6.8
Pending MTF VM Exits
As noted in Section 23.5.2, VM entry may cause an MTF VM exit to be pending immediately after VM entry. The following items detail the treatment of these VM exits:
•
System-management interrupts (SMIs), INIT signals, and higher priority events
take priority over these VM exits. These VM exits take priority over debug-trap
exceptions and lower priority events.
•
These VM exits wake the logical processor if it just entered the HLT state because
of a VM entry (see Section 23.6.2). They do not occur if the logical processor just
entered the shutdown state or the wait-for-SIPI state.
23.6.9
VM Entries and Advanced Debugging Features
VM entries are not logged with last-branch records, do not produce branch-trace
messages, and do not update the branch-trace store.
23.7
VM-ENTRY FAILURES DURING OR AFTER LOADING
GUEST STATE
VM-entry failures due to the checks identified in Section 23.3.1 and failures during
the MSR loading identified in Section 23.4 are treated differently from those that
occur earlier in VM entry. In these cases, the following steps take place:
1. Information about the VM-entry failure is recorded in the VM-exit information
fields:
— Exit reason.
•
Bits 15:0 of this field contain the basic exit reason. It is loaded with a
number indicating the general cause of the VM-entry failure. The
following numbers are used:
33. VM-entry failure due to invalid guest state. A VM entry failed one of
the checks identified in Section 23.3.1.
34. VM-entry failure due to MSR loading. A VM entry failed in an attempt
to load MSRs (see Section 23.4).
41. VM-entry failure due to machine check. A machine check occurred
during VM entry (see Section 23.8).
23-34 Vol. 3B
VM ENTRIES
•
•
Bit 31 is set to 1 to indicate a VM-entry failure.
The remainder of the field (bits 30:16) is cleared.
— Exit qualification. This field is set based on the exit reason.
•
VM-entry failure due to invalid guest state. In most cases, the exit qualification is cleared to 0. The following non-zero values are used in the
cases indicated:
1. Not used.
2. Failure was due to a problem loading the PDPTEs (see Section
23.3.1.6).
3. Failure was due to an attempt to inject a non-maskable interrupt
(NMI) into a guest that is blocking events through the STI blocking bit
in the interruptibility-state field. Such failures are implementationspecific (see Section 23.3.1.5).
4. Failure was due to an invalid VMCS link pointer (see Section
23.3.1.5).
VM-entry checks on guest-state fields may be performed in any order.
Thus, an indication by exit qualification of one cause does not imply that
there are not also other errors. Different processors may give different
exit qualifications for the same VMCS.
•
VM-entry failure due to MSR loading. The exit qualification is loaded to
indicate which entry in the VM-entry MSR-load area caused the problem
(1 for the first entry, 2 for the second, etc.).
— All other VM-exit information fields are unmodified.
2. Processor state is loaded as would be done on a VM exit (see Section 24.5). If
this results in [CR4.PAE & CR0.PG & ~IA32_EFER.LMA] = 1, page-directorypointer-table entries (PDPTEs) may be checked and loaded (see Section 24.5.4).
3. The state of blocking by NMI is what it was before VM entry.
4. MSRs are loaded as specified in the VM-exit MSR-load area (see Section 24.6).
Although this process resembles that of a VM exit, many steps taken during a VM exit
do not occur for these VM-entry failures:
•
•
•
•
Most VM-exit information fields are not updated (see step 1 above).
The valid bit in the VM-entry interruption-information field is not cleared.
The guest-state area is not modified.
No MSRs are saved into the VM-exit MSR-store area.
23.8
MACHINE CHECKS DURING VM ENTRY
If a machine check occurs during a VM entry, one of the following occurs:
Vol. 3B 23-35
VM ENTRIES
•
The machine check is handled normally:
— If CR4.MCE = 1, a machine-check exception (#MC) is delivered through the
IDT.
— If CR4.MCE = 0, operation of the logical processor depends on whether the
logical processor is in SMX operation:1
•
•
If the logical processor is in SMX operation, an Intel® TXT shutdown
condition occurs. The error code used is 000CH, indicating “unrecoverable
machine check condition.” See Intel® Trusted Execution Technology
Preliminary Architecture Specification.
•
If the logical processor is outside SMX operation, it goes to the shutdown
state.
A VM-entry failure occurs as described in Section 23.7. The basic exit reason is
41, for “VM-entry failure due to machine check.”
The first option is not used if the machine check occurs after any guest state has
been loaded.
1. A logical processor is in SMX operation if GETSEC[SEXIT] has not been executed since the last
execution of GETSEC[SENTER]. A logical processor is outside SMX operation if GETSEC[SENTER]
has not been executed or if GETSEC[SEXIT] was executed after the last execution of GETSEC[SENTER]. See Chapter 6, “Safer Mode Extensions Reference,” in Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 2B.
23-36 Vol. 3B
CHAPTER 24
VM EXITS
VM exits occur in response to certain instructions and events in VMX non-root operation. Section 22.1 through Section 22.3 detail the causes of VM exits. VM exits
perform the following operation:
1. Information about the cause of the VM exit is recorded in the VM-exit information
fields and VM-entry control fields are modified as described in Section 24.2.
2. Processor state is saved in the guest-state area (Section 24.3).
3. MSRs may be saved in the VM-exit MSR-store area (Section 24.4).
4. The following may be performed in parallel and in any order (Section 24.5):
— Processor state is loaded based in part on the host-state area and some
VM-exit controls. This step is not performed for SMM VM exits that activate
the dual-monitor treatment of SMIs and SMM. See Section 26.15.6 for
information on how processor state is loaded by such VM exits.
— Address-range monitoring is cleared.
5. MSRs may be loaded from the VM-exit MSR-load area (Section 24.6). This step is
not performed for SMM VM exits that activate the dual-monitor treatment of
SMIs and SMM.
VM exits are not logged with last-branch records, do not produce branch-trace
messages, and do not update the branch-trace store.
Section 24.1 clarifies the nature of the architectural state before a VM exit begins.
The steps described above are detailed in Section 24.2 through Section 24.6.
Section 26.15 describes the dual-monitor treatment of system-management interrupts (SMIs) and system-management mode (SMM). Under this treatment, ordinary
transitions to SMM are replaced by VM exits to a separate SMM monitor. Called SMM
VM exits, these are caused by the arrival of an SMI or the execution of VMCALL in
VMX root operation. SMM VM exits differ from other VM exits in ways that are
detailed in Section 26.15.2.
24.1
ARCHITECTURAL STATE BEFORE A VM EXIT
This section describes the architectural state that exists before a VM exit, especially
for VM exits caused by events that would normally be delivered through the IDT.
Note the following:
•
An exception causes a VM exit directly if the bit corresponding to that exception
is set in the exception bitmap. A non-maskable interrupt (NMI) causes a VM exit
directly if the “NMI exiting” VM-execution control is 1. An external interrupt
Vol. 3B 24-1
VM EXITS
causes a VM exit directly if the “external-interrupt exiting” VM-execution control
is 1. A start-up IPI (SIPI) that arrives while a logical processor is in the wait-forSIPI activity state causes a VM exit directly. INIT signals that arrive while the
processor is not in the wait-for-SIPI activity state cause VM exits directly.
•
An exception, NMI, external interrupt, or software interrupt causes a VM exit
indirectly if it does not do so directly but delivery of the event causes a nested
exception, double fault, task switch, APIC access (see Section 22.2), EPT
violation, or EPT misconfiguration that causes a VM exit.
•
An event results in a VM exit if it causes a VM exit (directly or indirectly).
The following bullets detail when architectural state is and is not updated in response
to VM exits:
•
If an event causes a VM exit directly, it does not update architectural state as it
would have if it had it not caused the VM exit:
— A debug exception does not update DR6, DR7.GD, or IA32_DEBUGCTL.LBR.
(Information about the nature of the debug exception is saved in the exit
qualification field.)
— A page fault does not update CR2. (The linear address causing the page fault
is saved in the exit-qualification field.)
— An NMI causes subsequent NMIs to be blocked, but only after the VM exit
completes.
— An external interrupt does not acknowledge the interrupt controller and the
interrupt remains pending, unless the “acknowledge interrupt on exit”
VM-exit control is 1. In such a case, the interrupt controller is acknowledged
and the interrupt is no longer pending.
— The flags L0 – L3 in DR7 (bit 0, bit 2, bit 4, and bit 6) are not cleared when a
task switch causes a VM exit.
— If a task switch causes a VM exit, none of the following are modified by the
task switch: old task-state segment (TSS); new TSS; old TSS descriptor; new
TSS descriptor; RFLAGS.NT1; or the TR register.
— No last-exception record is made if the event that would do so directly causes
a VM exit.
— If a machine-check exception causes a VM exit directly, this does not prevent
machine-check MSRs from being updated. These are updated by the machine
check itself and not the resulting machine-check exception.
1. This chapter uses the notation RAX, RIP, RSP, RFLAGS, etc. for processor registers because most
processors that support VMX operation also support Intel 64 architecture. For processors that do
not support Intel 64 architecture, this notation refers to the 32-bit forms of those registers
(EAX, EIP, ESP, EFLAGS, etc.). In a few places, notation such as EAX is used to refer specifically to
lower 32 bits of the indicated register.
24-2 Vol. 3B
VM EXITS
— If the logical processor is in an inactive state (see Section 21.4.2) and not
executing instructions, some events may be blocked but others may return
the logical processor to the active state. Unblocked events may cause
VM exits.1 If an unblocked event causes a VM exit directly, a return to the
active state occurs only after the VM exit completes.2 The VM exit generates
any special bus cycle that is normally generated when the active state is
entered from that activity state.
MTF VM exits (see Section 22.7.2 and Section 23.6.8) are not blocked in the
HLT activity state. If an MTF VM exit occurs in the HLT activity state, the
logical processor returns to the active state only after the VM exit completes.
MTF VM exits are blocked the shutdown state and the wait-for-SIPI state.
•
If an event causes a VM exit indirectly, the event does update architectural state:
— A debug exception updates DR6, DR7, and the IA32_DEBUGCTL MSR. No
debug exceptions are considered pending.
— A page fault updates CR2.
— An NMI causes subsequent NMIs to be blocked before the VM exit
commences.
— An external interrupt acknowledges the interrupt controller and the interrupt
is no longer pending.
— If the logical processor had been in an inactive state, it enters the active state
and, before the VM exit commences, generates any special bus cycle that is
normally generated when the active state is entered from that activity state.
— There is no blocking by STI or by MOV SS when the VM exit commences.
— Processor state that is normally updated as part of delivery through the IDT
(CS, RIP, SS, RSP, RFLAGS) is not modified. However, the incomplete delivery
of the event may write to the stack.
— The treatment of last-exception records is implementation dependent:
•
Some processors make a last-exception record when beginning the
delivery of an event through the IDT (before it can encounter a nested
exception). Such processors perform this update even if the event
encounters a nested exception that causes a VM exit (including the case
where nested exceptions lead to a triple fault).
•
Other processors delay making a last-exception record until event
delivery has reached some event handler successfully (perhaps after one
or more nested exceptions). Such processors do not update the last-
1. If a VM exit takes the processor from an inactive state resulting from execution of a specific
instruction (HLT or MWAIT), the value saved for RIP by that VM exit will reference the following
instruction.
2. An exception is made if the logical processor had been inactive due to execution of MWAIT; in
this case, it is considered to have become active before the VM exit.
Vol. 3B 24-3
VM EXITS
exception record if a VM exit or triple fault occurs before an event handler
is reached.
•
If the “virtual NMIs” VM-execution control is 1, VM entry injects an NMI, and
delivery of the NMI causes a nested exception, double fault, task switch, or APIC
access that causes a VM exit, virtual-NMI blocking is in effect before the VM exit
commences.
•
If a VM exit results from a fault, EPT violation, or EPT misconfiguration
encountered during execution of IRET and the “NMI exiting” VM-execution
control is 0, any blocking by NMI is cleared before the VM exit commences.
However, the previous state of blocking by NMI may be recorded in the VM-exit
interruption-information field; see Section 24.2.2.
•
If a VM exit results from a fault, EPT violation, or EPT misconfiguration
encountered during execution of IRET and the “virtual NMIs” VM-execution
control is 1, virtual-NMI blocking is cleared before the VM exit commences.
However, the previous state of virtual-NMI blocking may be recorded in the
VM-exit interruption-information field; see Section 24.2.2.
•
Suppose that a VM exit is caused directly by an x87 FPU Floating-Point Error
(#MF) or by any of the following events if the event was unblocked due to (and
given priority over) an x87 FPU Floating-Point Error: an INIT signal, an external
interrupt, an NMI, an SMI; or a machine-check exception. In these cases, there
is no blocking by STI or by MOV SS when the VM exit commences.
•
Normally, a last-branch record may be made when an event is delivered through
the IDT. However, if such an event results in a VM exit before delivery is
complete, no last-branch record is made.
•
If machine-check exception results in a VM exit, processor state is suspect and
may result in suspect state being saved to the guest-state area. A VM monitor
should consult the RIPV and EIPV bits in the IA32_MCG_STATUS MSR before
resuming a guest that caused a VM exit resulting from a machine-check
exception.
•
If a VM exit results from a fault, APIC access (see Section 22.2), EPT violation, or
EPT misconfiguration encountered while executing an instruction, data
breakpoints due to that instruction may have been recognized and information
about them may be saved in the pending debug exceptions field (see Section
24.3.4).
•
The following VM exits are considered to happen after an instruction is executed:
— VM exits resulting from debug traps (single-step, I/O breakpoints, and data
breakpoints).
— VM exits resulting from debug exceptions whose recognition was delayed by
blocking by MOV SS.
— VM exits resulting from some machine-check exceptions.
— Trap-like VM exits due to execution of MOV to CR8 when the “CR8-load
exiting” VM-execution control is 0 and the “use TPR shadow” VM-execution
24-4 Vol. 3B
VM EXITS
control is 1. (Such VM exits can occur only from 64-bit mode and thus only on
processors that support Intel 64 architecture.)
— Trap-like VM exits due to execution of WRMSR when the “use MSR bitmaps”
VM-execution control is 1, the value of ECX is 808H, bit 808H in write bitmap
for low MSRs is 0, and the “virtualize x2APIC mode” VM-execution control is
1. See Section 22.1.3.
— VM exits caused by TPR-shadow updates (see Section 22.5.3.3) that result
from APIC accesses as part of instruction execution.
For these VM exits, the instruction’s modifications to architectural state complete
before the VM exit occurs. Such modifications include those to the logical
processor’s interruptibility state (see Table 21-3). If there had been blocking by
MOV SS, POP SS, or STI before the instruction executed, such blocking is no
longer in effect.
24.2
RECORDING VM-EXIT INFORMATION AND UPDATING
VM-ENTRY CONTROL FIELDS
VM exits begin by recording information about the nature of and reason for the
VM exit in the VM-exit information fields. Section 24.2.1 to Section 24.2.4 detail the
use of these fields.
In addition to updating the VM-exit information fields, the valid bit (bit 31) is cleared
in the VM-entry interruption-information field. If bit 5 of the IA32_VMX_MISC MSR
(index 485H) is read as 1 (see Appendix G.6), the value of IA32_EFER.LMA is stored
into the “IA-32e mode guest” VM-entry control.1
24.2.1
Basic VM-Exit Information
Section 21.9.1 defines the basic VM-exit information fields. The following items detail
their use.
•
Exit reason.
— Bits 15:0 of this field contain the basic exit reason. It is loaded with a number
indicating the general cause of the VM exit. Appendix I lists the numbers used
and their meaning.
— The remainder of the field (bits 31:16) is cleared to 0 (certain SMM VM exits
may set some of these bits; see Section 26.15.2.3).2
1. Bit 5 of the IA32_VMX_MISC MSR is read as 1 on any logical processor that supports the 1-setting of the “unrestricted guest” VM-execution control.
2. Bit 13 of this field is set on certain VM-entry failures; see Section 23.7.
Vol. 3B 24-5
VM EXITS
•
Exit qualification. This field is saved for VM exits due to the following causes:
debug exceptions; page-fault exceptions; start-up IPIs (SIPIs); systemmanagement interrupts (SMIs) that arrive immediately after the retirement of
I/O instructions; task switches; INVEPT; INVLPG; INVVPID; LGDT; LIDT; LLDT;
LTR; SGDT; SIDT; SLDT; STR; VMCLEAR; VMPTRLD; VMPTRST; VMREAD;
VMWRITE; VMXON; control-register accesses; MOV DR; I/O instructions;
MWAIT; accesses to the APIC-access page (see Section 22.2); and EPT violations.
For all other VM exits, this field is cleared. The following items provide details:
— For a debug exception, the exit qualification contains information about the
debug exception. The information has the format given in Table 24-1.
Table 24-1. Exit Qualification for Debug Exceptions
Bit Position(s)
Contents
3:0
B3 – B0. When set, each of these bits indicates that the corresponding
breakpoint condition was met. Any of these bits may be set even if its
corresponding enabling bit in DR7 is not set.
12:4
Reserved (cleared to 0).
13
BD. When set, this bit indicates that the cause of the debug exception is
“debug register access detected.”
14
BS. When set, this bit indicates that the cause of the debug exception is
either the execution of a single instruction (if RFLAGS.TF = 1 and
IA32_DEBUGCTL.BTF = 0) or a taken branch (if
RFLAGS.TF = DEBUGCTL.BTF = 1).
63:15
Reserved (cleared to 0). Bits 63:32 exist only on processors that
support Intel 64 architecture.
— For a page-fault exception, the exit qualification contains the linear address
that caused the page fault. On processors that support Intel 64 architecture,
bits 63:32 are cleared if the logical processor was not in 64-bit mode before
the VM exit.
— For a start-up IPI (SIPI), the exit qualification contains the SIPI vector
information in bits 7:0. Bits 63:8 of the exit qualification are cleared to 0.
— For a task switch, the exit qualification contains details about the task switch,
encoded as shown in Table 24-2.
Table 24-2. Exit Qualification for Task Switch
Bit Position(s)
Contents
15:0
Selector of task-state segment (TSS) to which the guest attempted to switch
24-6 Vol. 3B
VM EXITS
Table 24-2. Exit Qualification for Task Switch (Contd.)
Bit Position(s)
Contents
29:16
Reserved (cleared to 0)
31:30
Source of task switch initiation:
0: CALL instruction
1: IRET instruction
2: JMP instruction
3: Task gate in IDT
63:32
Reserved (cleared to 0). These bits exist only on processors that support Intel
64 architecture.
— For INVLPG, the exit qualification contains the linear-address operand of the
instruction.
•
On processors that support Intel 64 architecture, bits 63:32 are cleared if
the logical processor was not in 64-bit mode before the VM exit.
•
If the INVLPG source operand specifies an unusable segment, the linear
address specified in the exit qualification will match the linear address
that the INVLPG would have used if no VM exit occurred. This address is
not architecturally defined and may be implementation-specific.
— For INVEPT, INVVPID, LGDT, LIDT, LLDT, LTR, SGDT, SIDT, SLDT, STR,
VMCLEAR, VMPTRLD, VMPTRST, VMREAD, VMWRITE, and VMXON, the exit
qualification receives the value of the instruction’s displacement field, which
is sign-extended to 64 bits if necessary (32 bits on processors that do not
support Intel 64 architecture). If the instruction has no displacement (for
example, has a register operand), zero is stored into the exit qualification.
On processors that support Intel 64 architecture, an exception is made for
RIP-relative addressing (used only in 64-bit mode). Such addressing causes
an instruction to use an address that is the sum of the displacement field
and the value of RIP that references the following instruction. In this case,
the exit qualification is loaded with the sum of the displacement field and
the appropriate RIP value.
In all cases, bits of this field beyond the instruction’s address size are
undefined. For example, suppose that the address-size field in the VM-exit
instruction-information field (see Section 21.9.4 and Section 24.2.4) reports
an n-bit address size. Then bits 63:n (bits 31:n on processors that do not
support Intel 64 architecture) of the instruction displacement are undefined.
Vol. 3B 24-7
VM EXITS
— For a control-register access, the exit qualification contains information about
the access and has the format given in Table 24-3.
Table 24-3. Exit Qualification for Control-Register Accesses
Bit Positions
Contents
3:0
Number of control register (0 for CLTS and LMSW). Bit 3 is always 0 on
processors that do not support Intel 64 architecture as they do not support CR8.
5:4
Access type:
0 = MOV to CR
1 = MOV from CR
2 = CLTS
3 = LMSW
6
LMSW operand type:
0 = register
1 = memory
For CLTS and MOV CR, cleared to 0
7
Reserved (cleared to 0)
11:8
For MOV CR, the general-purpose register:
0 = RAX
1 = RCX
2 = RDX
3 = RBX
4 = RSP
5 = RBP
6 = RSI
7 = RDI
8–15 represent R8–R15, respectively (used only on processors that support
Intel 64 architecture)
For CLTS and LMSW, cleared to 0
15:12
31:16
Reserved (cleared to 0)
For LMSW, the LMSW source data
For CLTS and MOV CR, cleared to 0
63:32
24-8 Vol. 3B
Reserved (cleared to 0). These bits exist only on processors that support Intel
64 architecture.
VM EXITS
— For MOV DR, the exit qualification contains information about the instruction
and has the format given in Table 24-4.
Table 24-4. Exit Qualification for MOV DR
Bit Position(s)
Contents
2:0
Number of debug register
3
Reserved (cleared to 0)
4
Direction of access (0
7:5
Reserved (cleared to 0)
11:8
General-purpose register:
= MOV to DR; 1 = MOV from DR)
0 = RAX
1 = RCX
2 = RDX
3 = RBX
4 = RSP
5 = RBP
6 = RSI
7 = RDI
8 –15 = R8 – R15, respectively
63:12
Reserved (cleared to 0)
— For an I/O instruction, the exit qualification contains information about the
instruction and has the format given in Table 24-5.
Table 24-5. Exit Qualification for I/O Instructions
Bit Position(s)
Contents
2:0
Size of access:
0 = 1-byte
1 = 2-byte
3 = 4-byte
Other values not used
= OUT, 1 = IN)
3
Direction of the attempted access (0
4
String instruction (0 = not string; 1 = string)
Vol. 3B 24-9
VM EXITS
Table 24-5. Exit Qualification for I/O Instructions (Contd.)
Bit Position(s)
Contents
5
REP prefixed (0 = not REP; 1 = REP)
6
Operand encoding (0 = DX, 1 = immediate)
15:7
Reserved (cleared to 0)
31:16
Port number (as specified in DX or in an immediate operand)
63:32
Reserved (cleared to 0). These bits exist only on processors that support Intel
64 architecture.
— For MWAIT, the exit qualification contains a value that indicates whether
address-range monitoring hardware was armed. The exit qualification is set
either to 0 (if address-range monitoring hardware is not armed) or to 1 (if
address-range monitoring hardware is armed).
— For an APIC-access VM exit resulting from a linear access or a guest-physical
access to the APIC-access page (see Section 22.2.1 and Section 22.2.2), the
exit qualification contains information about the access and has the format
given in Table 24-6.1
Table 24-6. Exit Qualification for APIC-Access VM Exits from Linear Accesses and
Guest-Physical Accesses
Bit Position(s)
Contents
11:0
• If the APIC-access VM exit is due to a linear access, the offset of access
within the APIC page.
• Undefined if the APIC-access VM exit is due a guest-physical access
15:12
Access type:
0 = linear access for a data read during instruction execution
1 = linear access for a data write during instruction execution
2 = linear access for an instruction fetch
3 = linear access (read or write) during event delivery
10 = guest-physical access during event delivery
15 = guest-physical access for an instruction fetch or during instruction
execution
Other values not used
63:16
24-10 Vol. 3B
Reserved (cleared to 0). Bits 63:32 exist only on processors that support
Intel 64 architecture.
VM EXITS
Such a VM exit that set bits 15:12 of the exit qualification to 0000b (data
read during instruction execution) or 0001b (data write during instruction
execution) set bit 12—which distinguishes data read from data write—to that
which would have been stored in bit 1—W/R—of the page-fault error code had
the access caused a page fault instead of an APIC-access VM exit. This
implies the following:
•
For an APIC-access VM exit caused by the CLFLUSH instruction, the
access type is “data read during instruction execution.”
•
For an APIC-access VM exit caused by the ENTER instruction, the access
type is “data write during instruction execution.”
•
For an APIC-access VM exit caused by the MASKMOVQ instruction or the
MASKMOVDQU instruction, the access type is “data write during
instruction execution.”
•
For an APIC-access VM exit caused by the MONITOR instruction, the
access type is “data read during instruction execution.”
Such a VM exit stores 1 for bit 31 for IDT-vectoring information field (see
Section 24.2.3) if and only if it sets bits 15:12 of the exit qualification to
0011b (linear access during event delivery) or 1010b (guest-physical access
during event delivery).
See Section 22.2.1.3 for further discussion of these instructions and APICaccess VM exits.
For APIC-access VM exits resulting from physical accesses, the APIC-access
page (see Section 22.2.3), the exit qualification is undefined.
— For an EPT violation, the exit qualification contains information about the
access causing the EPT violation and has the format given in Table 24-5.
Table 24-7. Exit Qualification for EPT Violations
Bit Position(s)
Contents
0
Set if the access causing the EPT violation was a data read.
1
Set if the access causing the EPT violation was a data write.
2
Set if the access causing the EPT violation was an instruction fetch.
1. The exit qualification is undefined if the access was part of the logging of a branch record or a
precise-event-based-sampling (PEBS) record to the DS save area. It is recommended that software configure the paging structures so that no address in the DS save area translates to an
address on the APIC-access page.
Vol. 3B 24-11
VM EXITS
Table 24-7. Exit Qualification for EPT Violations (Contd.)
Bit Position(s)
Contents
3
The logical-AND of bit 0 in the EPT paging-structures entries used to translate
the guest-physical address of the access causing the EPT violation (indicates
that the guest-physical address was readable).1
4
The logical-AND of bit 1 in the EPT paging-structures entries used to translate
the guest-physical address of the access causing the EPT violation (indicates
that the guest-physical address was writeable).
5
The logical-AND of bit 2 in the EPT paging-structures entries used to translate
the guest-physical address of the access causing the EPT violation (indicates
that the guest-physical address was executable).
6
Reserved (cleared to 0).
7
Set if the guest linear-address field is valid.
The guest linear-address field is valid for all EPT violations except those
resulting from an attempt to load the guest PDPTEs as part of the execution of
the MOV CR instruction.
8
If bit 7 is 1:
• Set if the access causing the EPT violation is to a guest-physical address
that is the translation of a linear address.
• Clear if the access causing the EPT violation is to a paging-structure entry
as part of a page walk or the update of an accessed or dirty bit.
Reserved if bit 7 is 0 (cleared to 0).
11:9
Reserved (cleared to 0).
12
NMI unblocking due to IRET
63:13
Reserved (cleared to 0).
NOTES:
1. Bits 5:3 are cleared to 0 if any of EPT paging-structures entries used to translate the guest-physical address of the access causing the EPT violation is not present (see Section 25.2.2).
An EPT violation that occurs during as a result of execution of a read-modifywrite operation sets bit 1 (data write). Whether it also sets bit 0 (data read)
is implementation-specific and, for a given implementation, may differ for
different kinds of read-modify-write operations.
Bit 12 is undefined in any of the following cases:
•
If the “NMI exiting” VM-execution control is 1 and the “virtual NMIs”
VM-execution control is 0.
24-12 Vol. 3B
VM EXITS
•
If the VM exit sets the valid bit in the IDT-vectoring information field (see
Section 24.2.3).
Otherwise, bit 12 is defined as follows:
•
•
If the “virtual NMIs” VM-execution control is 0, the EPT violation was
caused by a memory access as part of execution of the IRET instruction,
and blocking by NMI (see Table 21-3) was in effect before execution of
IRET, bit 12 is set to 1.
•
If the “virtual NMIs” VM-execution control is 1,the EPT violation was
caused by a memory access as part of execution of the IRET instruction,
and virtual-NMI blocking was in effect before execution of IRET, bit 12 is
set to 1.
•
For all other relevant VM exits, bit 12 is cleared to 0.
Guest-linear address. For some VM exits, this field receives a linear address
that pertains to the VM exit. The field is set for different VM exits as follows:
— VM exits due to attempts to execute LMSW with a memory operand. In these
cases, this field receives the linear address of that operand. Bits 63:32 are
cleared if the logical processor was not in 64-bit mode before the VM exit.
— VM exits due to attempts to execute INS or OUTS for which the relevant
segment is usable (if the relevant segment is not usable, the value is
undefined). (ES is always the relevant segment for INS; for OUTS, the
relevant segment is DS unless overridden by an instruction prefix.) The linear
address is the base address of relevant segment plus (E)DI (for INS) or (E)SI
(for OUTS). Bits 63:32 are cleared if the logical processor was not in 64-bit
mode before the VM exit.
— VM exits due to EPT violations that set bit 7 of the exit qualification (see
Table 24-7; these are all EPT violations except those resulting from an
attempt to load the PDPTEs as of execution of the MOV CR instruction). The
linear address may translate to the guest-physical address whose access
caused the EPT violation. Alternatively, translation of the linear address may
reference a paging-structure entry whose access caused the EPT violation.
Bits 63:32 are cleared if the logical processor was not in 64-bit mode before
the VM exit.
— For all other VM exits, the field is undefined.
•
Guest-physical address. For a VM exit due to an EPT violation or an EPT
misconfiguration, this field receives the guest-physical address that caused the
EPT violation or EPT misconfiguration. For all other VM exits, the field is
undefined.
24.2.2
Information for VM Exits Due to Vectored Events
Section 21.9.2 defines fields containing information for VM exits due to the following
events: exceptions (including those generated by the instructions INT3, INTO,
Vol. 3B 24-13
VM EXITS
BOUND, and UD2); external interrupts that occur while the “acknowledge interrupt
on exit” VM-exit control is 1; and non-maskable interrupts (NMIs). Such VM exits
include those that occur on an attempt at a task switch that causes an exception
before generating the VM exit due to the task switch that causes the VM exit.
The following items detail the use of these fields:
•
VM-exit interruption information (format given in Table 21-14). The following
items detail how this field is established for VM exits due to these events:
— For an exception, bits 7:0 receive the exception vector (at most 31). For an
NMI, bits 7:0 are set to 2. For an external interrupt, bits 7:0 receive the
interrupt number.
— Bits 10:8 are set to 0 (external interrupt), 2 (non-maskable interrupt), 3
(hardware exception), or 6 (software exception). Hardware exceptions
comprise all exceptions except breakpoint exceptions (#BP; generated by
INT3) and overflow exceptions (#OF; generated by INTO); these are
software exceptions. BOUND-range exceeded exceptions (#BR; generated by
BOUND) and invalid opcode exceptions (#UD) generated by UD2 are
hardware exceptions.
— Bit 11 is set to 1 if the VM exit is caused by a hardware exception that would
have delivered an error code on the stack. This bit is always 0 if the VM exit
occurred while the logical processor was in real-address mode (CR0.PE=0).1
If bit 11 is set to 1, the error code is placed in the VM-exit interruption error
code (see below).
— Bit 12 is undefined in any of the following cases:
•
If the “NMI exiting” VM-execution control is 1 and the “virtual NMIs”
VM-execution control is 0.
•
If the VM exit sets the valid bit in the IDT-vectoring information field (see
Section 24.2.3).
•
If the VM exit is due to a double fault (the interruption type is hardware
exception and the vector is 8).
Otherwise, bit 12 is defined as follows:
•
If the “virtual NMIs” VM-execution control is 0, the VM exit is due to a
fault on the IRET instruction (other than a debug exception for an
instruction breakpoint), and blocking by NMI (see Table 21-3) was in
effect before execution of IRET, bit 12 is set to 1.
•
If the “virtual NMIs” VM-execution control is 1, the VM exit is due to a
fault on the IRET instruction (other than a debug exception for an
1. If the capability MSR IA32_VMX_CR0_FIXED0 reports that CR0.PE must be 1 in VMX operation, a
logical processor cannot be in real-address mode unless the “unrestricted guest” VM-execution
control and bit 31 of the primary processor-based VM-execution controls are both 1.
24-14 Vol. 3B
VM EXITS
instruction breakpoint), and virtual-NMI blocking was in effect before
execution of IRET, bit 12 is set to 1.
•
For all other relevant VM exits, bit 12 is cleared to 0.1
— Bits 30:13 are always set to 0.
— Bit 31 is always set to 1.
For other VM exits (including those due to external interrupts when the
“acknowledge interrupt on exit” VM-exit control is 0), the field is marked invalid
(by clearing bit 31) and the remainder of the field is undefined.
•
VM-exit interruption error code.
— For VM exits that set both bit 31 (valid) and bit 11 (error code valid) in the
VM-exit interruption-information field, this field receives the error code that
would have been pushed on the stack had the event causing the VM exit been
delivered normally through the IDT. The EXT bit is set in this field exactly
when it would be set normally. For exceptions that occur during the delivery
of double fault (if the IDT-vectoring information field indicates a double fault),
the EXT bit is set to 1, assuming that (1) that the exception would produce an
error code normally (if not incident to double-fault delivery) and (2) that the
error code uses the EXT bit (not for page faults, which use a different format).
— For other VM exits, the value of this field is undefined.
24.2.3
Information for VM Exits During Event Delivery
Section 21.9.3 defined fields containing information for VM exits that occur while
delivering an event through the IDT and as a result of any of the following cases:2
•
A fault occurs during event delivery and causes a VM exit (because the bit
associated with the fault is set to 1 in the exception bitmap).
•
A task switch is invoked through a task gate in the IDT. The VM exit occurs due to
the task switch only after the initial checks of the task switch pass (see Section
22.6.2).
•
•
Event delivery causes an APIC-access VM exit (see Section 22.2).
An EPT violation or EPT misconfiguration that occurs during event delivery.
These fields are used for VM exits that occur during delivery of events injected as
part of VM entry (see Section 23.5.1.2).
1. The conditions imply that, if the “NMI exiting” VM-execution control is 0 or the “virtual NMIs” VMexecution control is 1, bit 12 is always cleared to 0 by VM exits due to debug exceptions.
2. This includes the case in which a VM exit occurs while delivering a software interrupt (INT n)
through the 16-bit IVT (interrupt vector table) that is used in virtual-8086 mode with virtualmachine extensions (if RFLAGS.VM = CR4.VME = 1).
Vol. 3B 24-15
VM EXITS
A VM exit is not considered to occur during event delivery in any of the following
circumstances:
•
The original event causes the VM exit directly (for example, because the original
event is a non-maskable interrupt (NMI) and the “NMI exiting” VM-execution
control is 1).
•
The original event results in a double-fault exception that causes the VM exit
directly.
•
The VM exit occurred as a result of fetching the first instruction of the handler
invoked by the event delivery.
•
The VM exit is caused by a triple fault.
The following items detail the use of these fields:
•
IDT-vectoring information (format given in Table 21-15). The following items
detail how this field is established for VM exits that occur during event delivery:
— If the VM exit occurred during delivery of an exception, bits 7:0 receive the
exception vector (at most 31). If the VM exit occurred during delivery of an
NMI, bits 7:0 are set to 2. If the VM exit occurred during delivery of an
external interrupt, bits 7:0 receive the interrupt number.
— Bits 10:8 are set to indicate the type of event that was being delivered when
the VM exit occurred: 0 (external interrupt), 2 (non-maskable interrupt), 3
(hardware exception), 4 (software interrupt), 5 (privileged software
interrupt), or 6 (software exception).
Hardware exceptions comprise all exceptions except breakpoint exceptions
(#BP; generated by INT3) and overflow exceptions (#OF; generated by
INTO); these are software exceptions. BOUND-range exceeded exceptions
(#BR; generated by BOUND) and invalid opcode exceptions (#UD) generated
by UD2 are hardware exceptions.
Bits 10:8 may indicate privileged software interrupt if such an event was
injected as part of VM entry.
— Bit 11 is set to 1 if the VM exit occurred during delivery of a hardware
exception that would have delivered an error code on the stack. This bit is
always 0 if the VM exit occurred while the logical processor was in realaddress mode (CR0.PE=0).1 If bit 11 is set to 1, the error code is placed in
the IDT-vectoring error code (see below).
— Bit 12 is undefined.
— Bits 30:13 are always set to 0.
— Bit 31 is always set to 1.
1. If the capability MSR IA32_VMX_CR0_FIXED0 reports that CR0.PE must be 1 in VMX operation, a
logical processor cannot be in real-address mode unless the “unrestricted guest” VM-execution
control and bit 31 of the primary processor-based VM-execution controls are both 1.
24-16 Vol. 3B
VM EXITS
For other VM exits, the field is marked invalid (by clearing bit 31) and the
remainder of the field is undefined.
•
IDT-vectoring error code.
— For VM exits that set both bit 31 (valid) and bit 11 (error code valid) in the
IDT-vectoring information field, this field receives the error code that would
have been pushed on the stack by the event that was being delivered through
the IDT at the time of the VM exit. The EXT bit is set in this field when it would
be set normally.
— For other VM exits, the value of this field is undefined.
24.2.4
Information for VM Exits Due to Instruction Execution
Section 21.9.4 defined fields containing information for VM exits that occur due to
instruction execution. (The VM-exit instruction length is also used for VM exits that
occur during the delivery of a software interrupt or software exception.) The
following items detail their use.
•
VM-exit instruction length. This field is used in the following cases:
— For fault-like VM exits due to attempts to execute one of the following
instructions that cause VM exits unconditionally (see Section 22.1.2) or
based on the settings of VM-execution controls (see Section 22.1.3): CLTS,
CPUID, GETSEC, HLT, IN, INS, INVD, INVEPT, INVLPG, INVVPID, LGDT, LIDT,
LLDT, LMSW, LTR, MONITOR, MOV CR, MOV DR, MWAIT, OUT, OUTS, PAUSE,
RDMSR, RDPMC, RDTSC, RDTSCP, RSM, SGDT, SIDT, SLDT, STR, VMCALL,
VMCLEAR, VMLAUNCH, VMPTRLD, VMPTRST, VMREAD, VMRESUME,
VMWRITE, VMXOFF, VMXON, WBINVD, WRMSR, and XSETBV.1
— For VM exits due to software exceptions (those generated by executions of
INT3 or INTO).
— For VM exits due to faults encountered during delivery of a software
interrupt, privileged software exception, or software exception.
— For VM exits due to attempts to effect a task switch via instruction execution.
These are VM exits that produce an exit reason indicating task switch and
either of the following:
•
An exit qualification indicating execution of CALL, IRET, or JMP
instruction.
•
An exit qualification indicating a task gate in the IDT and an IDT-vectoring
information field indicating that the task gate was encountered during
1. This item applies only to fault-like VM exits. It does not apply to trap-like VM exits following executions of the MOV to CR8 instruction when the “use TPR shadow” VM-execution control is 1 or
to those following executions of the WRMSR instruction when the “virtualize x2APIC mode” VMexecution control is 1.
Vol. 3B 24-17
VM EXITS
delivery of a software interrupt, privileged software exception, or
software exception.
— For APIC-access VM exits resulting from linear accesses (see Section 22.2.1)
and encountered during delivery of a software interrupt, privileged software
exception, or software exception.1
In all the above cases, this field receives the length in bytes (1–15) of the
instruction (including any instruction prefixes) whose execution led to the
VM exit (see the next paragraph for one exception).
The cases of VM exits encountered during delivery of a software interrupt,
privileged software exception, or software exception include those encountered
during delivery of events injected as part of VM entry (see Section 23.5.1.2). If
the original event was injected as part of VM entry, this field receives the value of
the VM-entry instruction length.
All VM exits other than those listed in the above items leave this field undefined.
•
VM-exit instruction information. For VM exits due to attempts to execute
INS, INVEPT, INVVPID, LIDT, LGDT, LLDT, LTR, OUTS, SIDT, SGDT, SLDT, STR,
VMCLEAR, VMPTRLD, VMPTRST, VMREAD, VMWRITE, or VMXON, this field
receives information about the instruction that caused the VM exit. The format of
the field depends on the identity of the instruction causing the VM exit:
— For VM exits due to attempts to execute INS or OUTS, the field has the format
is given in Table 24-8.2
Table 24-8. Format of the VM-Exit Instruction-Information Field as Used for INS and
OUTS
Bit Position(s) Content
6:0
Undefined.
9:7
Address size:
0: 16-bit
1: 32-bit
2: 64-bit (used only on processors that support Intel 64 architecture)
Other values not used.
14:10
Undefined.
1. The VM-exit instruction-length field is not defined following APIC-access VM exits resulting from
physical accesses (see Section 22.2.3) even if encountered during delivery of a software interrupt, privileged software exception, or software exception.
2. The format of the field was undefined for these VM exits on the first processors to support the
virtual-machine extensions. Software can determine whether the format specified in Table 24-8
is used by consulting the VMX capability MSR IA32_VMX_BASIC (see Appendix G.1).
24-18 Vol. 3B
VM EXITS
Table 24-8. Format of the VM-Exit Instruction-Information Field as Used for INS and
OUTS (Contd.)
Bit Position(s) Content
17:15
Segment register:
0: ES
1: CS
2: SS
3: DS
4: FS
5: GS
Other values not used. Undefined for VM exits due to execution of INS.
31:18
Undefined.
— For VM exits due to attempts to execute LIDT, LGDT, SIDT, or SGDT, the field
has the format is given in Table 24-9.
Table 24-9. Format of the VM-Exit Instruction-Information Field as Used for LIDT,
LGDT, SIDT, or SGDT
Bit Position(s) Content
1:0
Scaling:
0: no scaling
1: scale by 2
2: scale by 4
3: scale by 8 (used only on processors that support Intel 64 architecture)
Undefined for instructions with no index register (bit 22 is set).
6:2
Undefined.
9:7
Address size:
0: 16-bit
1: 32-bit
2: 64-bit (used only on processors that support Intel 64 architecture)
Other values not used.
10
Cleared to 0.
11
Operand size:
0: 16-bit
1: 32-bit
Undefined for VM exits from 64-bit mode.
14:12
Undefined.
Vol. 3B 24-19
VM EXITS
Table 24-9. Format of the VM-Exit Instruction-Information Field as Used for LIDT,
LGDT, SIDT, or SGDT (Contd.)
Bit Position(s) Content
17:15
Segment register:
0: ES
1: CS
2: SS
3: DS
4: FS
5: GS
Other values not used.
21:18
IndexReg:
0 = RAX
1 = RCX
2 = RDX
3 = RBX
4 = RSP
5 = RBP
6 = RSI
7 = RDI
8–15 represent R8–R15, respectively (used only on processors that support
Intel 64 architecture)
Undefined for instructions with no index register (bit 22 is set).
22
IndexReg invalid (0 = valid; 1 = invalid)
26:23
BaseReg (encoded as IndexReg above)
Undefined for instructions with no base register (bit 27 is set).
27
BaseReg invalid (0 = valid; 1 = invalid)
29:28
Instruction identity:
0: SGDT
1: SIDT
2: LGDT
3: LIDT
31:30
Undefined.
— For VM exits due to attempts to execute LLDT, LTR, SLDT, or STR, the field has
the format is given in Table 24-10.
— For VM exits due to attempts to execute VMCLEAR, VMPTRLD, VMPTRST, or
VMXON, the field has the format is given in Table 24-11.
— For VM exits due to attempts to execute VMREAD or VMWRITE, the field has
the format is given in Table 24-12.
24-20 Vol. 3B
VM EXITS
Table 24-10. Format of the VM-Exit Instruction-Information Field as Used for LLDT,
LTR, SLDT, and STR
Bit Position(s) Content
1:0
Scaling:
0: no scaling
1: scale by 2
2: scale by 4
3: scale by 8 (used only on processors that support Intel 64 architecture)
Undefined for register instructions (bit 10 is set) and for memory instructions with
no index register (bit 10 is clear and bit 22 is set).
2
Undefined.
6:3
Reg1:
0 = RAX
1 = RCX
2 = RDX
3 = RBX
4 = RSP
5 = RBP
6 = RSI
7 = RDI
8–15 represent R8–R15, respectively (used only on processors that support
Intel 64 architecture)
Undefined for memory instructions (bit 10 is clear).
9:7
Address size:
0: 16-bit
1: 32-bit
2: 64-bit (used only on processors that support Intel 64 architecture)
Other values not used. Undefined for register instructions (bit 10 is set).
10
Mem/Reg (0 = memory; 1 = register).
14:11
Undefined.
17:15
Segment register:
0: ES
1: CS
2: SS
3: DS
4: FS
5: GS
Other values not used. Undefined for register instructions (bit 10 is set).
Vol. 3B 24-21
VM EXITS
Table 24-10. Format of the VM-Exit Instruction-Information Field as Used for LLDT,
LTR, SLDT, and STR (Contd.)
Bit Position(s) Content
21:18
IndexReg (encoded as Reg1 above)
Undefined for register instructions (bit 10 is set) and for memory instructions with
no index register (bit 10 is clear and bit 22 is set).
22
IndexReg invalid (0 = valid; 1 = invalid)
Undefined for register instructions (bit 10 is set).
26:23
BaseReg (encoded as Reg1 above)
Undefined for register instructions (bit 10 is set) and for memory instructions with
no base register (bit 10 is clear and bit 27 is set).
27
BaseReg invalid (0 = valid; 1 = invalid)
Undefined for register instructions (bit 10 is set).
29:28
Instruction identity:
0: SLDT
1: STR
2: LLDT
3: LTR
31:30
Undefined.
Table 24-11. Format of the VM-Exit Instruction-Information Field as Used for
VMCLEAR, VMPTRLD, VMPTRST, and VMXON
Bit Position(s) Content
1:0
Scaling:
0: no scaling
1: scale by 2
2: scale by 4
3: scale by 8 (used only on processors that support Intel 64 architecture)
Undefined for instructions with no index register (bit 22 is set).
6:2
Undefined.
9:7
Address size:
0: 16-bit
1: 32-bit
2: 64-bit (used only on processors that support Intel 64 architecture)
Other values not used.
10
Cleared to 0.
14:11
Undefined.
24-22 Vol. 3B
VM EXITS
Table 24-11. Format of the VM-Exit Instruction-Information Field as Used for
VMCLEAR, VMPTRLD, VMPTRST, and VMXON (Contd.)
Bit Position(s) Content
17:15
Segment register:
0: ES
1: CS
2: SS
3: DS
4: FS
5: GS
Other values not used.
21:18
IndexReg:
0 = RAX
1 = RCX
2 = RDX
3 = RBX
4 = RSP
5 = RBP
6 = RSI
7 = RDI
8–15 represent R8–R15, respectively (used only on processors that support
Intel 64 architecture)
Undefined for instructions with no index register (bit 22 is set).
22
IndexReg invalid (0 = valid; 1 = invalid)
26:23
BaseReg (encoded as IndexReg above)
27
BaseReg invalid (0 = valid; 1 = invalid)
31:28
Undefined.
Undefined for instructions with no base register (bit 27 is set).
Table 24-12. Format of the VM-Exit Instruction-Information Field as Used for
VMREAD and VMWRITE
Bit Position(s) Content
1:0
Scaling:
0: no scaling
1: scale by 2
2: scale by 4
3: scale by 8 (used only on processors that support Intel 64 architecture)
Undefined for register instructions (bit 10 is set) and for memory instructions with
no index register (bit 10 is clear and bit 22 is set).
Vol. 3B 24-23
VM EXITS
Table 24-12. Format of the VM-Exit Instruction-Information Field as Used for
VMREAD and VMWRITE (Contd.)
Bit Position(s) Content
2
Undefined.
6:3
Reg1:
0 = RAX
1 = RCX
2 = RDX
3 = RBX
4 = RSP
5 = RBP
6 = RSI
7 = RDI
8–15 represent R8–R15, respectively (used only on processors that support
Intel 64 architecture)
Undefined for memory instructions (bit 10 is clear).
9:7
Address size:
0: 16-bit
1: 32-bit
2: 64-bit (used only on processors that support Intel 64 architecture)
Other values not used. Undefined for register instructions (bit 10 is set).
10
Mem/Reg (0 = memory; 1 = register).
14:11
Undefined.
17:15
Segment register:
0: ES
1: CS
2: SS
3: DS
4: FS
5: GS
Other values not used. Undefined for register instructions (bit 10 is set).
21:18
IndexReg (encoded as Reg1 above)
Undefined for register instructions (bit 10 is set) and for memory instructions with
no index register (bit 10 is clear and bit 22 is set).
22
IndexReg invalid (0 = valid; 1 = invalid)
Undefined for register instructions (bit 10 is set).
24-24 Vol. 3B
VM EXITS
Table 24-12. Format of the VM-Exit Instruction-Information Field as Used for
VMREAD and VMWRITE (Contd.)
Bit Position(s) Content
26:23
BaseReg (encoded as Reg1 above)
Undefined for register instructions (bit 10 is set) and for memory instructions with
no base register (bit 10 is clear and bit 27 is set).
27
BaseReg invalid (0 = valid; 1 = invalid)
Undefined for register instructions (bit 10 is set).
31:28
Reg2 (same encoding as Reg1 above)
— For VM exits due to attempts to execute INVEPT or INVVPID, the field has the
format is given in Table 24-13.
Table 24-13. Format of the VM-Exit Instruction-Information Field as Used for INVEPT
and INVVPID
Bit Position(s) Content
1:0
Scaling:
0: no scaling
1: scale by 2
2: scale by 4
3: scale by 8 (used only on processors that support Intel 64 architecture)
Undefined for instructions with no index register (bit 22 is set).
6:2
Undefined.
9:7
Address size:
0: 16-bit
1: 32-bit
2: 64-bit (used only on processors that support Intel 64 architecture)
Other values not used.
10
Cleared to 0.
14:11
Undefined.
17:15
Segment register:
0: ES
1: CS
2: SS
3: DS
4: FS
5: GS
Other values not used.
Vol. 3B 24-25
VM EXITS
Table 24-13. Format of the VM-Exit Instruction-Information Field as Used for INVEPT
and INVVPID (Contd.)
Bit Position(s) Content
21:18
IndexReg:
0 = RAX
1 = RCX
2 = RDX
3 = RBX
4 = RSP
5 = RBP
6 = RSI
7 = RDI
8–15 represent R8–R15, respectively (used only on processors that support
Intel 64 architecture)
Undefined for instructions with no index register (bit 22 is set).
22
IndexReg invalid (0 = valid; 1 = invalid)
26:23
BaseReg (encoded as IndexReg above)
Undefined for memory instructions with no base register (bit 27 is set).
27
BaseReg invalid (0 = valid; 1 = invalid)
31:28
Reg2 (same encoding as IndexReg above)
For all other VM exits, the field is undefined.
•
I/O RCX, I/O RSI, I/O RDI, I/O RIP. These fields are undefined except for
SMM VM exits due to system-management interrupts (SMIs) that arrive
immediately after retirement of I/O instructions. See Section 26.15.2.3.
24.3
SAVING GUEST STATE
Each field in the guest-state area of the VMCS (see Section 21.4) is written with the
corresponding component of processor state. On processors that support Intel 64
architecture, the full values of each natural-width field (see Section 21.10.2) is saved
regardless of the mode of the logical processor before and after the VM exit.
In general, the state saved is that which was in the logical processor at the time the
VM exit commences. See Section 24.1 for a discussion of which architectural updates
occur at that time.
Section 24.3.1 through Section 24.3.4 provide details for how certain components of
processor state are saved. These sections reference VMCS fields that correspond to
processor state. Unless otherwise stated, these references are to fields in the gueststate area.
24-26 Vol. 3B
VM EXITS
24.3.1
Saving Control Registers, Debug Registers, and MSRs
Contents of certain control registers, debug registers, and MSRs is saved as follows:
•
The contents of CR0, CR3, CR4, and the IA32_SYSENTER_CS,
IA32_SYSENTER_ESP, and IA32_SYSENTER_EIP MSRs are saved into the corresponding fields. Bits 63:32 of the IA32_SYSENTER_CS MSR are not saved. On
processors that do not support Intel 64 architecture, bits 63:32 of the
IA32_SYSENTER_ESP and IA32_SYSENTER_EIP MSRs are not saved.
•
If the “save debug controls” VM-exit control is 1, the contents of DR7 and the
IA32_DEBUGCTL MSR are saved into the corresponding fields. The first
processors to support the virtual-machine extensions supported only the 1setting of this control and thus always saved data into these fields.
•
If the “save IA32_PAT” VM-exit control is 1, the contents of the IA32_PAT MSR
are saved into the corresponding field.
•
If the “save IA32_EFER” VM-exit control is 1, the contents of the IA32_EFER MSR
are saved into the corresponding field.
•
The value of the SMBASE field is undefined after all VM exits except SMM
VM exits. See Section 26.15.2.
24.3.2
Saving Segment Registers and Descriptor-Table Registers
For each segment register (CS, SS, DS, ES, FS, GS, LDTR, or TR), the values saved
for the base-address, segment-limit, and access rights are based on whether the
register was unusable (see Section 21.4.1) before the VM exit:
•
If the register was unusable, the values saved into the following fields are
undefined: (1) base address; (2) segment limit; and (3) bits 7:0 and bits 15:12
in the access-rights field. The following exceptions apply:
— CS.
•
•
The base-address and segment-limit fields are saved.
The L, D, and G bits are saved in the access-rights field.
— SS.
•
•
DPL is saved in the access-rights field.
On processors that support Intel 64 architecture, bits 63:32 of the value
saved for the base address are always zero.
— DS and ES. On processors that support Intel 64 architecture, bits 63:32 of
the values saved for the base addresses are always zero.
— FS and GS. The base-address field is saved.
— LDTR. The value saved for the base address is always canonical.
Vol. 3B 24-27
VM EXITS
•
If the register was not unusable, the values saved into the following fields are
those which were in the register before the VM exit: (1) base address;
(2) segment limit; and (3) bits 7:0 and bits 15:12 in access rights.
•
Bits 31:17 and 11:8 in the access-rights field are always cleared. Bit 16 is set to
1 if and only if the segment is unusable.
The contents of the GDTR and IDTR registers are saved into the corresponding baseaddress and limit fields.
24.3.3
Saving RIP, RSP, and RFLAGS
The contents of the RIP, RSP, and RFLAGS registers are saved as follows:
•
The value saved in the RIP field is determined by the nature and cause of the
VM exit:
— If the VM exit occurs due to by an attempt to execute an instruction that
causes VM exits unconditionally or that has been configured to cause a
VM exit via the VM-execution controls, the value saved references that
instruction.
— If the VM exit is caused by an occurrence of an INIT signal, a start-up IPI
(SIPI), or system-management interrupt (SMI), the value saved is that which
was in RIP before the event occurred.
— If the VM exit occurs due to the 1-setting of either the “interrupt-window
exiting” VM-execution control or the “NMI-window exiting” VM-execution
control, the value saved is that which would be in the register had the VM exit
not occurred.
— If the VM exit is due to an external interrupt, non-maskable interrupt (NMI),
or hardware exception (as defined in Section 24.2.2), the value saved is the
return pointer that would have been saved (either on the stack had the event
been delivered through a trap or interrupt gate,1 or into the old task-state
segment had the event been delivered through a task gate).
— If the VM exit is due to a triple fault, the value saved is the return pointer that
would have been saved (either on the stack had the event been delivered
through a trap or interrupt gate, or into the old task-state segment had the
event been delivered through a task gate) had delivery of the double fault not
encountered the nested exception that caused the triple fault.
— If the VM exit is due to a software exception (due to an execution of INT3 or
INTO), the value saved references the INT3 or INTO instruction that caused
that exception.
— Suppose that the VM exit is due to a task switch that was caused by execution
of CALL, IRET, or JMP or by execution of a software interrupt (INT n) or
1. The reference here is to the full value of RIP before any truncation that would occur had the
stack width been only 32 bits or 16 bits.
24-28 Vol. 3B
VM EXITS
software exception (due to execution of INT3 or INTO) that encountered a
task gate in the IDT. The value saved references the instruction that caused
the task switch (CALL, IRET, JMP, INT n, INT3, or INTO).
— Suppose that the VM exit is due to a task switch that was caused by a task
gate in the IDT that was encountered for any reason except the direct access
by a software interrupt or software exception. The value saved is that which
would have been saved in the old task-state segment had the task switch
completed normally.
— If the VM exit is due to an execution of MOV to CR8 or WRMSR that reduced
the value of the TPR shadow1 below that of TPR threshold VM-execution
control field, the value saved references the instruction following the MOV to
CR8 or WRMSR.
— If the VM exit was caused by a TPR-shadow update (see Section 21.5.3.3)
that results from an APIC access as part of instruction execution, the value
saved references the instruction following the one whose execution caused
the VTPR access.
•
•
The contents of the RSP register are saved into the RSP field.
With the exception of the resume flag (RF; bit 16), the contents of the RFLAGS
register is saved into the RFLAGS field. RFLAGS.RF is saved as follows:
— If the VM exit is caused directly by an event that would normally be delivered
through the IDT, the value saved is that which would appear in the saved
RFLAGS image (either that which would be saved on the stack had the event
been delivered through a trap or interrupt gate2 or into the old task-state
segment had the event been delivered through a task gate) had the event
been delivered through the IDT. See below for VM exits due to task switches
caused by task gates in the IDT.
— If the VM exit is caused by a triple fault, the value saved is that which the
logical processor would have in RF in the RFLAGS register had the triple fault
taken the logical processor to the shutdown state.
— If the VM exit is caused by a task switch (including one caused by a task gate
in the IDT), the value saved is that which would have been saved in the
RFLAGS image in the old task-state segment (TSS) had the task switch
completed normally without exception.
— If the VM exit is caused by an attempt to execute an instruction that unconditionally causes VM exits or one that was configured to do with a VM-execution
control, the value saved is 0.3
1. The TPR shadow is bits 7:4 of the byte at offset 80H of the virtual-APIC page (see Section
21.6.8).
2. The reference here is to the full value of RFLAGS before any truncation that would occur had the
stack width been only 32 bits or 16 bits.
Vol. 3B 24-29
VM EXITS
— For APIC-access VM exits and for VM exits caused by EPT violations and EPT
misconfigurations, the value saved depends on whether the VM exit occurred
during delivery of an event through the IDT:
•
If the VM exit stored 0 for bit 31 for IDT-vectoring information field
(because the VM exit did not occur during delivery of an event through
the IDT; see Section 24.2.3), the value saved is 1.
•
If the VM exit stored 1 for bit 31 for IDT-vectoring information field
(because the VM exit did occur during delivery of an event through the
IDT), the value saved is the value that would have appeared in the saved
RFLAGS image had the event been delivered through the IDT (see
above).
— For all other VM exits, the value saved is the value RFLAGS.RF had before the
VM exit occurred.
24.3.4
Saving Non-Register State
Information corresponding to guest non-register state is saved as follows:
•
The activity-state field is saved with the logical processor’s activity state before
the VM exit.1 See Section 24.1 for details of how events leading to a VM exit may
affect the activity state.
•
The interruptibility-state field is saved to reflect the logical processor’s interruptibility before the VM exit. See Section 24.1 for details of how events leading to a
VM exit may affect this state. VM exits that end outside system-management
mode (SMM) save bit 2 (blocking by SMI) as 0 regardless of the state of such
blocking before the VM exit.
Bit 3 (blocking by NMI) is treated specially if the “virtual NMIs” VM-execution
control is 1. In this case, the value saved for this field does not indicate the
blocking of NMIs but rather the state of virtual-NMI blocking.
•
The pending debug exceptions field is saved as clear for all VM exits except the
following:
— A VM exit caused by an INIT signal, a machine-check exception, or a systemmanagement interrupt (SMI).
— A VM exit with basic exit reason either “TPR below threshold.”2
— A VM exit with basic exit reason “monitor trap flag.”
3. This is true even if RFLAGS.RF was 1 before the instruction was executed. If, in response to such
a VM exit, a VM monitor re-enters the guest to re-execute the instruction that caused the
VM exit (for example, after clearing the VM-execution control that caused the VM exit), the
instruction may encounter a code breakpoint that has already been processed. A VM monitor can
avoid this by setting the guest value of RFLAGS.RF to 1 before resuming guest software.
1. If this activity state was an inactive state resulting from execution of a specific instruction (HLT
or MWAIT), the value saved for RIP by that VM exit will reference the following instruction.
24-30 Vol. 3B
VM EXITS
— VM exits that are not caused by debug exceptions and that occur while there
is MOV-SS blocking of debug exceptions.
For VM exits that do not clear the field, the value saved is determined as follows:
— Each of bits 3:0 may be set if it corresponds to a matched breakpoint. This
may be true even if the corresponding breakpoint is not enabled in DR7.
— Suppose that a VM exit is due to an INIT signal, a machine-check exception,
or an SMI; or that a VM exit has basic exit reason “TPR below threshold” or
“monitor trap flag.” In this case, the value saved sets bits corresponding to
the causes of any debug exceptions that were pending at the time of the
VM exit.
If the VM exit occurs immediately after VM entry, the value saved may match
that which was loaded on VM entry (see Section 23.6.3). Otherwise, the
following items apply:
•
Bit 12 (enabled breakpoint) is set to 1 if there was at least one matched
data or I/O breakpoint that was enabled in DR7. Bit 12 is also set if it had
been set on VM entry, causing there to be valid pending debug exceptions
(see Section 23.6.3) and the VM exit occurred before those exceptions
were either delivered or lost. In other cases, bit 12 is cleared to 0.
•
Bit 14 (BS) is set if RFLAGS.TF = 1 in either of the following cases:
•
IA32_DEBUGCTL.BTF = 0 and the cause of a pending debug
exception was the execution of a single instruction.
•
IA32_DEBUGCTL.BTF = 1 and the cause of a pending debug
exception was a taken branch.
— Suppose that a VM exit is due to another reason (but not a debug exception)
and occurs while there is MOV-SS blocking of debug exceptions. In this case,
the value saved sets bits corresponding to the causes of any debug
exceptions that were pending at the time of the VM exit. If the VM exit occurs
immediately after VM entry (no instructions were executed in VMX non-root
operation), the value saved may match that which was loaded on VM entry
(see Section 23.6.3). Otherwise, the following items apply:
•
Bit 12 (enabled breakpoint) is set to 1 if there was at least one matched
data or I/O breakpoint that was enabled in DR7. Bit 12 is also set if it had
been set on VM entry, causing there to be valid pending debug exceptions
(see Section 23.6.3) and the VM exit occurred before those exceptions
were either delivered or lost. In other cases, bit 12 is cleared to 0.
•
The setting of bit 14 (BS) is implementation-specific. However, it is not
set if RFLAGS.TF = 0 or IA32_DEBUGCTL.BTF = 1.
— The reserved bits in the field are cleared.
2. This item includes VM exits that occur after executions of MOV to CR8 or WRMSR (Section
22.1.3), TPR-shadow updates (Section 22.5.3.3), and certain VM entries (Section 23.6.7).
Vol. 3B 24-31
VM EXITS
•
If the “save VMX-preemption timer value” VM-exit control is 1, the value of timer
is saved into the VMX-preemption timer-value field. This is the value loaded from
this field on VM entry as subsequently decremented (see Section 22.7.1). VM
exits due to timer expiration save the value 0. Other VM exits may also save the
value 0 if the timer expired during VM exit. (If the “save VMX-preemption timer
value” VM-exit control is 0, VM exit does not modify the value of the VMXpreemption timer-value field.)
•
If the logical processor supports the 1-setting of the “enable EPT” VM-execution
control, values are saved into the four (4) PDPTE fields as follows:
— If the “enable EPT” VM-execution control is 1 and the logical processor was
using PAE paging at the time of the VM exit, the PDPTE values currently in use
are saved:1
•
•
•
The values saved into bits 11:9 of each of the fields is undefined.
If the value saved into one of the fields has bit 0 (present) clear, the value
saved into bits 63:1 of that field is undefined. That value need not
correspond to the value that was loaded by VM entry or to any value that
might have been loaded in VMX non-root operation.
If the value saved into one of the fields has bit 0 (present) set, the value
saved into bits 63:12 of the field is a guest-physical address.
— If the “enable EPT” VM-execution control is 0 or the logical processor was not
using PAE paging at the time of the VM exit, the values saved are undefined.
24.4
SAVING MSRS
After processor state is saved to the guest-state area, values of MSRs may be stored
into the VM-exit MSR-store area (see Section 21.7.2). Specifically each entry in that
area (up to the number specified in the VM-exit MSR-store count) is processed in
order by storing the value of the MSR indexed by bits 31:0 (as they would be read by
RDMSR) into bits 127:64. Processing of an entry fails in either of the following cases:
•
The value of bits 31:8 is 000008H, meaning that the indexed MSR is one that
allows access to an APIC register when the local APIC is in x2APIC mode.
•
The value of bits 31:0 indicates an MSR that can be read only in systemmanagement mode (SMM) and the VM exit will not end in SMM.
•
The value of bits 31:0 indicates an MSR that cannot be saved on VM exits for
model-specific reasons. A processor may prevent certain MSRs (based on the
1. A logical processor uses PAE paging if CR0.PG = 1, CR4.PAE = 1 and IA32_EFER.LMA = 0. See
Section 4.4 in the Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 3A.
“Enable EPT” is a secondary processor-based VM-execution control. If bit 31 of the primary processor-based VM-execution controls is 0, VM exit functions as if the “enable EPT” VM-execution
control were 0. See Section 21.6.2.
24-32 Vol. 3B
VM EXITS
value of bits 31:0) from being stored on VM exits, even if they can normally be
read by RDMSR. Such model-specific behavior is documented in Appendix B.
•
•
Bits 63:32 of the entry are not all 0.
An attempt to read the MSR indexed by bits 31:0 would cause a generalprotection exception if executed via RDMSR with CPL = 0.
A VMX abort occurs if processing fails for any entry. See Section 24.7.
24.5
LOADING HOST STATE
Processor state is updated on VM exits in the following ways:
•
Some state is loaded from or otherwise determined by the contents of the hoststate area.
•
•
•
Some state is determined by VM-exit controls.
Some state is established in the same way on every VM exit.
The page-directory pointers are loaded based on the values of certain control
registers.
This loading may be performed in any order.
On processors that support Intel 64 architecture, the full values of each 64-bit field
loaded (for example, the base address for GDTR) is loaded regardless of the mode of
the logical processor before and after the VM exit.
The loading of host state is detailed in Section 24.5.1 to Section 24.5.5. These
sections reference VMCS fields that correspond to processor state. Unless otherwise
stated, these references are to fields in the host-state area.
A logical processor is in IA-32e mode after a VM exit only if the “host address-space
size” VM-exit control is 1. If the logical processor was in IA-32e mode before the
VM exit and this control is 0, a VMX abort occurs. See Section 24.7.
In addition to loading host state, VM exits clear address-range monitoring (Section
24.5.6).
After the state loading described in this section, VM exits may load MSRs from the
VM-exit MSR-load area (see Section 24.6). This loading occurs only after the state
loading described in this section.
24.5.1
Loading Host Control Registers, Debug Registers, MSRs
VM exits load new values for controls registers, debug registers, and some MSRs:
•
CR0, CR3, and CR4 are loaded from the CR0 field, the CR3 field, and the CR4
field, respectively, with the following exceptions:
— The following bits are not modified:
Vol. 3B 24-33
VM EXITS
•
For CR0, ET, CD, NW; bits 63:32 (on processors that support Intel 64
architecture), 28:19, 17, and 15:6; and any bits that are fixed in VMX
operation (see Section 20.8).1
•
For CR3, bits 63:52 and bits in the range 51:32 beyond the processor’s
physical-address width (they are cleared to 0).2 (This item applies only to
processors that support Intel 64 architecture.)
•
For CR4, any bits that are fixed in VMX operation (see Section 20.8).
— CR4.PAE is set to 1 if the “host address-space size” VM-exit control is 1.
— CR4.PCIDE is set to 0 if the “host address-space size” VM-exit control is 0.
•
•
DR7 is set to 400H.
The following MSRs are established as follows:
— The IA32_DEBUGCTL MSR is cleared to 00000000_00000000H.
— The IA32_SYSENTER_CS MSR is loaded from the IA32_SYSENTER_CS field.
Since that field has only 32 bits, bits 63:32 of the MSR are cleared to 0.
— IA32_SYSENTER_ESP MSR and IA32_SYSENTER_EIP MSR are loaded from
the IA32_SYSENTER_ESP field and the IA32_SYSENTER_EIP field, respectively.
If the processor does not support the Intel 64 architecture, these fields have
only 32 bits; bits 63:32 of the MSRs are cleared to 0.
If the processor does support the Intel 64 architecture and the processor
supports N < 64 linear-address bits, each of bits 63:N is set to the value of
bit N–1.3
— The following steps are performed on processors that support Intel 64 architecture:
•
The MSRs FS.base and GS.base are loaded from the base-address fields
for FS and GS, respectively (see Section 24.5.2).
•
The LMA and LME bits in the IA32_EFER MSR are each loaded with the
setting of the “host address-space size” VM-exit control.
— If the “load IA32_PERF_GLOBAL_CTRL” VM-exit control is 1, the
IA32_PERF_GLOBAL_CTRL MSR is loaded from the
IA32_PERF_GLOBAL_CTRL field. Bits that are reserved in that MSR are
maintained with their reserved values.
1. Bits 28:19, 17, and 15:6 of CR0 and CR0.ET are unchanged by executions of MOV to CR0. CR0.ET
is always 1 and the other bits are always 0.
2. Software can determine a processor’s physical-address width by executing CPUID with
80000008H in EAX. The physical-address width is returned in bits 7:0 of EAX.
3. Software can determine the number N by executing CPUID with 80000008H in EAX. The number of linear-address bits supported is returned in bits 15:8 of EAX.
24-34 Vol. 3B
VM EXITS
— If the “load IA32_PAT” VM-exit control is 1, the IA32_PAT MSR is loaded from
the IA32_PAT field. Bits that are reserved in that MSR are maintained with
their reserved values.
— If the “load IA32_EFER” VM-exit control is 1, the IA32_EFER MSR is loaded
from the IA32_EFER field. Bits that are reserved in that MSR are maintained
with their reserved values.
With the exception of FS.base and GS.base, any of these MSRs is subsequently
overwritten if it appears in the VM-exit MSR-load area. See Section 24.6.
24.5.2
Loading Host Segment and Descriptor-Table Registers
Each of the registers CS, SS, DS, ES, FS, GS, and TR is loaded as follows (see below
for the treatment of LDTR):
•
The selector is loaded from the selector field. The segment is unusable if its
selector is loaded with zero. The checks specified Section 23.3.1.2 limit the
selector values that may be loaded. In particular, CS and TR are never loaded
with zero and are thus never unusable. SS can be loaded with zero only on
processors that support Intel 64 architecture and only if the VM exit is to 64-bit
mode (64-bit mode allows use of segments marked unusable).
•
The base address is set as follows:
— CS. Cleared to zero.
— SS, DS, and ES. Undefined if the segment is unusable; otherwise, cleared to
zero.
— FS and GS. Undefined (but, on processors that support Intel 64 architecture,
canonical) if the segment is unusable and the VM exit is not to 64-bit mode;
otherwise, loaded from the base-address field.
If the processor supports the Intel 64 architecture and the processor
supports N < 64 linear-address bits, each of bits 63:N is set to the value of
bit N–1.1 The values loaded for base addresses for FS and GS are also
manifest in the FS.base and GS.base MSRs.
— TR. Loaded from the host-state area. If the processor supports the Intel 64
architecture and the processor supports N < 64 linear-address bits, each of
bits 63:N is set to the value of bit N–1.
•
The segment limit is set as follows:
— CS. Set to FFFFFFFFH (corresponding to a descriptor limit of FFFFFH and a Gbit setting of 1).
— SS, DS, ES, FS, and GS. Undefined if the segment is unusable; otherwise, set
to FFFFFFFFH.
1. Software can determine the number N by executing CPUID with 80000008H in EAX. The number of linear-address bits supported is returned in bits 15:8 of EAX.
Vol. 3B 24-35
VM EXITS
— TR. Set to 00000067H.
•
The type field and S bit are set as follows:
— CS. Type set to 11 and S set to 1 (execute/read, accessed, non-conforming
code segment).
— SS, DS, ES, FS, and GS. Undefined if the segment is unusable; otherwise,
type set to 3 and S set to 1 (read/write, accessed, expand-up data segment).
— TR. Type set to 11 and S set to 0 (busy 32-bit task-state segment).
•
The DPL is set as follows:
— CS, SS, and TR. Set to 0. The current privilege level (CPL) will be 0 after the
VM exit completes.
— DS, ES, FS, and GS. Undefined if the segment is unusable; otherwise, set to
0.
•
The P bit is set as follows:
— CS, TR. Set to 1.
— SS, DS, ES, FS, and GS. Undefined if the segment is unusable; otherwise, set
to 1.
•
On processors that support Intel 64 architecture, CS.L is loaded with the setting
of the “host address-space size” VM-exit control. Because the value of this
control is also loaded into IA32_EFER.LMA (see Section 24.5.1), no VM exit is
ever to compatibility mode (which requires IA32_EFER.LMA = 1 and CS.L = 0).
•
D/B.
— CS. Loaded with the inverse of the setting of the “host address-space size”
VM-exit control. For example, if that control is 0, indicating a 32-bit guest,
CS.D/B is set to 1.
— SS. Set to 1.
— DS, ES, FS, and GS. Undefined if the segment is unusable; otherwise, set to
1.
— TR. Set to 0.
•
G.
— CS. Set to 1.
— SS, DS, ES, FS, and GS. Undefined if the segment is unusable; otherwise, set
to 1.
— TR. Set to 0.
The host-state area does not contain a selector field for LDTR. LDTR is established as
follows on all VM exits: the selector is cleared to 0000H, the segment is marked
unusable and is otherwise undefined (although the base address is always canonical).
24-36 Vol. 3B
VM EXITS
The base addresses for GDTR and IDTR are loaded from the GDTR base-address field
and the IDTR base-address field, respectively. If the processor supports the Intel 64
architecture and the processor supports N < 64 linear-address bits, each of bits 63:N
of each base address is set to the value of bit N–1 of that base address. The GDTR
and IDTR limits are each set to FFFFH.
24.5.3
Loading Host RIP, RSP, and RFLAGS
RIP and RSP are loaded from the RIP field and the RSP field, respectively. RFLAGS is
cleared, except bit 1, which is always set.
24.5.4
Checking and Loading Host Page-Directory-Pointer-Table
Entries
If CR0.PG = 1, CR4.PAE = 1, and IA32_EFER.LMA = 0, the logical processor uses
PAE paging. See Section 4.4 of the Intel® 64 and IA-32 Architectures Software
Developer’s Manual, Volume 3A.1 When in PAE paging is in use, the physical address
in CR3 references a table of page-directory-pointer-table entries (PDPTEs). A
MOV to CR3 when PAE paging is in use checks the validity of the PDPTEs and, if they
are valid, loads them into the processor (into internal, non-architectural registers).
A VM exit is to a VMM that uses PAE paging if (1) bit 5 (corresponding to CR4.PAE) is
set in the CR4 field in the host-state area of the VMCS; and (2) the “host addressspace size” VM-exit control is 0. Such a VM exit may check the validity of the PDPTEs
referenced by the CR3 field in the host-state area of the VMCS. Such a VM exit must
check their validity if either (1) PAE paging was not in use before the VM exit; or
(2) the value of CR3 is changing as a result of the VM exit. A VM exit to a VMM that
does not use PAE paging must not check the validity of the PDPTEs.
A VM exit that checks the validity of the PDPTEs uses the same checks that are used
when CR3 is loaded with MOV to CR3 when PAE paging is in use. If MOV to CR3 would
cause a general-protection exception due to the PDPTEs that would be loaded (e.g.,
because a reserved bit is set), a VMX abort occurs (see Section 24.7). If a VM exit to
a VMM that uses PAE does not cause a VMX abort, the PDPTEs are loaded into the
processor as would MOV to CR3, using the value of CR3 being load by the VM exit.
24.5.5
Updating Non-Register State
VM exits affect the non-register state of a logical processor as follows:
•
A logical processor is always in the active state after a VM exit.
1. On processors that support Intel 64 architecture, the physical-address extension may support
more than 36 physical-address bits. Software can determine a processor’s physical-address
width by executing CPUID with 80000008H in EAX. The physical-address width is returned in
bits 7:0 of EAX.
Vol. 3B 24-37
VM EXITS
•
Event blocking is affected as follows:
— There is no blocking by STI or by MOV SS after a VM exit.
— VM exits caused directly by non-maskable interrupts (NMIs) cause blocking
by NMI (see Table 21-3). Other VM exits do not affect blocking by NMI. (See
Section 24.1 for the case in which an NMI causes a VM exit indirectly.)
•
There are no pending debug exceptions after a VM exit.
Section 25.3 describes how the VMX architecture controls how a logical processor
manages information in the TLBs and paging-structure caches. The following items
detail how VM exits invalidate cached mappings:
•
If the “enable VPID” VM-execution control is 0, the logical processor invalidates
linear mappings and combined mappings associated with VPID 0000H (for all
PCIDs); combined mappings for VPID 0000H are invalidated for all EP4TA values
(EP4TA is the value of bits 51:12 of EPTP).
•
VM exits are not required to invalidate any guest-physical mappings, nor are they
required to invalidate any linear mappings or combined mappings if the “enable
VPID” VM-execution control is 1.
24.5.6
Clearing Address-Range Monitoring
The Intel 64 and IA-32 architectures allow software to monitor a specified address
range using the MONITOR and MWAIT instructions. See Section 8.10.4 in the Intel®
64 and IA-32 Architectures Software Developer’s Manual, Volume 3A. VM exits clear
any address-range monitoring that may be in effect.
24.6
LOADING MSRS
VM exits may load MSRs from the VM-exit MSR-load area (see Section 21.7.2).
Specifically each entry in that area (up to the number specified in the VM-exit MSRload count) is processed in order by loading the MSR indexed by bits 31:0 with the
contents of bits 127:64 as they would be written by WRMSR.
Processing of an entry fails in any of the following cases:
•
The value of bits 31:0 is either C0000100H (the IA32_FS_BASE MSR) or
C0000101H (the IA32_GS_BASE MSR).
•
The value of bits 31:8 is 000008H, meaning that the indexed MSR is one that
allows access to an APIC register when the local APIC is in x2APIC mode.
•
The value of bits 31:0 indicates an MSR that can be written only in systemmanagement mode (SMM) and the VM exit will not end in SMM.
(IA32_SMM_MONITOR_CTL is an MSR that can be written only in SMM.)
•
The value of bits 31:0 indicates an MSR that cannot be loaded on VM exits for
model-specific reasons. A processor may prevent loading of certain MSRs even if
24-38 Vol. 3B
VM EXITS
they can normally be written by WRMSR. Such model-specific behavior is
documented in Appendix B.
•
•
Bits 63:32 are not all 0.
An attempt to write bits 127:64 to the MSR indexed by bits 31:0 of the entry
would cause a general-protection exception if executed via WRMSR with
CPL = 0.1
If processing fails for any entry, a VMX abort occurs. See Section 24.7.
If any MSR is being loaded in such a way that would architecturally require a TLB
flush, the TLBs are updated so that, after VM exit, the logical processor does not use
any translations that were cached before the transition.
24.7
VMX ABORTS
A problem encountered during a VM exit leads to a VMX abort. A VMX abort takes a
logical processor into a shutdown state as described below.
A VMX abort does not modify the VMCS data in the VMCS region of any active VMCS.
The contents of these data are thus suspect after the VMX abort.
On a VMX abort, a logical processor saves a nonzero 32-bit VMX-abort indicator field
at byte offset 4 in the VMCS region of the VMCS whose misconfiguration caused the
failure (see Section 21.2). The following values are used:
1. There was a failure in saving guest MSRs (see Section 24.4).
2. Host checking of the page-directory-pointer-table entries (PDPTEs) failed (see
Section 24.5.4).
3. The current VMCS has been corrupted (through writes to the corresponding
VMCS region) in such a way that the logical processor cannot complete the
VM exit properly.
4. There was a failure on loading host MSRs (see Section 24.6).
5. There was a machine check during VM exit (see Section 24.8).
6. The logical processor was in IA-32e mode before the VM exit and the “host
address-space size” VM-entry control was 0 (see Section 24.5).
Some of these causes correspond to failures during the loading of state from the
host-state area. Because the loading of such state may be done in any order (see
Section 24.5) a VM exit that might lead to a VMX abort for multiple reasons (for
example, the current VMCS may be corrupt and the host PDPTEs might not be prop-
1. Note the following about processors that support Intel 64 architecture. If CR0.PG = 1, WRMSR to
the IA32_EFER MSR causes a general-protection exception if it would modify the LME bit. Since
CR0.PG is always 1 in VMX operation, the IA32_EFER MSR should not be included in the VM-exit
MSR-load area for the purpose of modifying the LME bit.
Vol. 3B 24-39
VM EXITS
erly configured). In such cases, the VMX-abort indicator could correspond to any one
of those reasons.
A logical processor never reads the VMX-abort indicator in a VMCS region and writes
it only with one of the non-zero values mentioned above. The VMX-abort indicator
allows software on one logical processor to diagnose the VMX-abort on another. For
this reason, it is recommended that software running in VMX root operation zero the
VMX-abort indicator in the VMCS region of any VMCS that it uses.
After saving the VMX-abort indicator, operation of a logical processor experiencing a
VMX abort depends on whether the logical processor is in SMX operation:1
•
If the logical processor is in SMX operation, an Intel® TXT shutdown condition
occurs. The error code used is 000DH, indicating “VMX abort.” See Intel® Trusted
Execution Technology Measured Launched Environment Programming Guide.
•
If the logical processor is outside SMX operation, it issues a special bus cycle (to
notify the chipset) and enters the VMX-abort shutdown state. RESET is the
only event that wakes a logical processor from the VMX-abort shutdown state.
The following events do not affect a logical processor in this state: machine
checks; INIT signals; external interrupts; non-maskable interrupts (NMIs); startup IPIs (SIPIs); and system-management interrupts (SMIs).
24.8
MACHINE CHECK DURING VM EXIT
If a machine check occurs during VM exit, one of the following occurs:
•
The machine check is handled normally:
— If CR4.MCE = 1, a machine-check exception (#MC) delivered through the
guest IDT.
— If CR4.MCE = 0, operation of the logical processor depends on whether the
logical processor is in SMX operation:2
•
If the logical processor is in SMX operation, an Intel® TXT shutdown
condition occurs. The error code used is 000CH, indicating “unrecoverable
machine check condition.” See Intel® Trusted Execution Technology
Measured Launched Environment Programming Guide.
1. A logical processor is in SMX operation if GETSEC[SEXIT] has not been executed since the last
execution of GETSEC[SENTER]. A logical processor is outside SMX operation if GETSEC[SENTER]
has not been executed or if GETSEC[SEXIT] was executed after the last execution of GETSEC[SENTER]. See Chapter 6, “Safer Mode Extensions Reference,” in Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 2B.
2. A logical processor is in SMX operation if GETSEC[SEXIT] has not been executed since the last
execution of GETSEC[SENTER]. A logical processor is outside SMX operation if GETSEC[SENTER]
has not been executed or if GETSEC[SEXIT] was executed after the last execution of GETSEC[SENTER]. See Chapter 6, “Safer Mode Extensions Reference,” in Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 2B.
24-40 Vol. 3B
VM EXITS
•
•
If the logical processor is outside SMX operation, it goes to the shutdown
state.
A VMX abort is generated (see Section 24.7). The logical processor blocks events
as done normally in VMX abort. The VMX abort indicator is 5, for “machine check
during VM exit.”
The first option is not used if the machine check occurs after any host state has been
loaded.
Vol. 3B 24-41
VM EXITS
24-42 Vol. 3B
CHAPTER 25
VMX SUPPORT FOR ADDRESS TRANSLATION
The architecture for VMX operation includes two features that support address translation: virtual-processor identifiers (VPIDs) and the extended page-table mechanism
(EPT). VPIDs are a mechanism for managing translations of linear addresses. EPT
defines a layer of address translation that augments the translation of linear
addresses.
Section 25.1 details the architecture of VPIDs. Section 25.2 provides the details of
EPT. Section 25.3 explains how a logical processor may cache information from the
paging structures, how it may use that cached information, and how software can
managed the cached information.
25.1
VIRTUAL PROCESSOR IDENTIFIERS (VPIDS)
The original architecture for VMX operation required VMX transitions to flush the TLBs
and paging-structure caches. This ensured that translations cached for the old linearaddress space would not be used after the transition.
Virtual-processor identifiers (VPIDs) introduce to VMX operation a facility by which
a logical processor may cache information for multiple linear-address spaces. When
VPIDs are used, VMX transitions may retain cached information and the logical
processor switches to a different linear-address space.
Section 25.3 details the mechanisms by which a logical processor manages information cached for multiple address spaces. A logical processor may tag some cached
information with a 16-bit VPID. This section specifies how the current VPID is determined at any point in time:
•
The current VPID is 0000H in the following situations:
— Outside VMX operation. (This includes operation in system-management
mode under the default treatment of SMIs and SMM with VMX operation; see
Section 26.14.)
— In VMX root operation.
— In VMX non-root operation when the “enable VPID” VM-execution control is 0.
•
If the logical processor is in VMX non-root operation and the “enable VPID” VMexecution control is 1, the current VPID is the value of the VPID VM-execution
control field in the VMCS. (VM entry ensures that this value is never 0000H; see
Section 23.2.1.1.)
VPIDs and PCIDs (see Section 4.10.1) can be used concurrently. When this is done,
the processor associates cached information with both a VPID and a PCID. Such
Vol. 3B 25-1
VMX SUPPORT FOR ADDRESS TRANSLATION
information is used only if the current VPID and PCID both match those associated
with the cached information.
25.2
THE EXTENDED PAGE TABLE MECHANISM (EPT)
The extended page-table mechanism (EPT) is a feature that can be used to support
the virtualization of physical memory. When EPT is in use, certain addresses that
would normally be treated as physical addresses (and used to access memory) are
instead treated as guest-physical addresses. Guest-physical addresses are translated by traversing a set of EPT paging structures to produce physical addresses
that are used to access memory.
•
•
•
•
Section 25.2.1 gives an overview of EPT.
Section 25.2.2 describes operation of EPT-based address translation.
Section 25.2.3 discusses VM exits that may be caused by EPT.
Section 25.2.4 describes interactions between EPT and memory typing.
25.2.1
EPT Overview
EPT is used when the “enable EPT” VM-execution control is 1.1 It translates the
guest-physical addresses used in VMX non-root operation and those used by
VM entry for event injection.
The translation from guest-physical addresses to physical addresses is determined
by a set of EPT paging structures. The EPT paging structures are similar to those
used to translate linear addresses while the processor is in IA-32e mode. Section
25.2.2 gives the details of the EPT paging structures.
If CR0.PG = 1, linear addresses are translated through paging structures referenced
through control register CR3 . While the “enable EPT” VM-execution control is 1,
these are called guest paging structures. There are no guest paging structures if
CR0.PG = 0.2
1. “Enable EPT” is a secondary processor-based VM-execution control. If bit 31 of the primary processor-based VM-execution controls is 0, the logical processor operates as if the “enable EPT”
VM-execution control were 0. See Section 21.6.2.
2. If the capability MSR IA32_VMX_CR0_FIXED0 reports that CR0.PG must be 1 in VMX operation,
CR0.PG can be 0 in VMX non-root operation only if the “unrestricted guest” VM-execution control
and bit 31 of the primary processor-based VM-execution controls are both 1.
25-2 Vol. 3B
VMX SUPPORT FOR ADDRESS TRANSLATION
When the “enable EPT” VM-execution control is 1, the identity of guest-physical
addresses depends on the value of CR0.PG:
•
•
If CR0.PG = 0, each linear address is treated as a guest-physical address.
If CR0.PG = 1, guest-physical addresses are those derived from the contents of
control register CR3 and the guest paging structures. (This includes the values of
the PDPTEs, which logical processors store in internal, non-architectural
registers.) The latter includes (in page-table entries and in other pagingstructure entries for which bit 7—PS—is 1) the addresses to which linear
addresses are translated by the guest paging structures.
If CR0.PG = 1, the translation of a linear address to a physical address requires
multiple translations of guest-physical addresses using EPT. Assume, for example,
that CR4.PAE = CR4.PSE = 0. The translation of a 32-bit linear address then operates as follows:
•
Bits 31:22 of the linear address select an entry in the guest page directory
located at the guest-physical address in CR3. The guest-physical address of the
guest page-directory entry (PDE) is translated through EPT to determine the
guest PDE’s physical address.
•
Bits 21:12 of the linear address select an entry in the guest page table located at
the guest-physical address in the guest PDE. The guest-physical address of the
guest page-table entry (PTE) is translated through EPT to determine the guest
PTE’s physical address.
•
Bits 11:0 of the linear address is the offset in the page frame located at the
guest-physical address in the guest PTE. The guest-physical address determined
by this offset is translated through EPT to determine the physical address to
which the original linear address translates.
In addition to translating a guest-physical address to a physical address, EPT specifies the privileges that software is allowed when accessing the address. Attempts at
disallowed accesses are called EPT violations and cause VM exits. See Section
25.2.3.
A logical processor uses EPT to translate guest-physical addresses only when those
addresses are used to access memory. This principle implies the following:
•
The MOV to CR3 instruction loads CR3 with a guest-physical address. Whether
that address is translated through EPT depends on whether PAE paging is being
used.1
— If PAE paging is not being used, the instruction does not use that address to
access memory and does not cause it to be translated through EPT. (If
CR0.PG = 1, the address will be translated through EPT on the next memory
accessing using a linear address.)
1. A logical processor uses PAE paging if CR0.PG = 1, CR4.PAE = 1 and IA32_EFER.LMA = 0. See
Section 4.4 in the Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 3A.
Vol. 3B 25-3
VMX SUPPORT FOR ADDRESS TRANSLATION
— If PAE paging is being used, the instruction loads the four (4) page-directorypointer-table entries (PDPTEs) from that address and it does cause the
address to be translated through EPT.
•
The MOV to CR0 instruction establishes PAE paging if it results in CR0.PG = 1 and
the following were held before the instruction executed: (1) CR0.PG = 0;
(2) CR4.PAE = 1; and (3) IA32_EFER.LME = 0. Such an execution loads the
PDPTEs from the guest-physical address in CR3. The address is translated
through EPT.
•
The MOV to CR4 instruction establishes PAE paging if it results in CR4.PAE = 1
and the following were held before the instruction executed: (1) CR0.PG = 1;
(2) CR4.PAE = 0; and (3) IA32_EFER.LMA = 0. Such an execution loads the
PDPTEs from the guest-physical address in CR3. The address is translated
through EPT.
•
The PDPTEs contain guest-physical addresses. The instructions that load the
PDPTEs (see above) do not use those addresses to access memory and do not
cause them to be translated through EPT. (The address in a PDPTE will be
translated through EPT on the next memory accessing using a linear address that
uses that PDPTE.)
25.2.2
EPT Translation Mechanism
The EPT translation mechanism uses only bits 47:0 of each guest-physical address.1
It uses a page-walk length of 4, meaning that at most 4 EPT paging-structure entries
are accessed to translate a guest-physical address.2
These 48 bits are partitioned by the logical processor to traverse the EPT paging
structures:
•
A 4-KByte naturally aligned EPT PML4 table is located at the physical address
specified in bits 51:12 of the extended-page-table pointer (EPTP), a VMexecution control field (see Table 21-8 in Section 21.6.11). An EPT PML4 table
comprises 512 64-bit entries (EPT PML4Es). An EPT PML4E is selected using the
physical address defined as follows:
— Bits 63:52 are all 0.
— Bits 51:12 are from the EPTP.
— Bits 11:3 are bits 47:39 of the guest-physical address.
1. No processors supporting the Intel 64 architecture support more than 48 physical-address bits.
Thus, no such processor can produce a guest-physical address with more than 48 bits. An
attempt to use such an address causes a page fault. An attempt to load CR3 with such an
address causes a general-protection fault. If PAE paging is being used, an attempt to load CR3
that would load a PDPTE with such an address causes a general-protection fault.
2. Future processors may include support for other EPT page-walk lengths. Software should read
the VMX capability MSR IA32_VMX_EPT_VPID_CAP (see Appendix G.10) to determine what EPT
page-walk lengths are supported.
25-4 Vol. 3B
VMX SUPPORT FOR ADDRESS TRANSLATION
— Bits 2:0 are all 0.
Because an EPT PML4E is identified using bits 47:39 of the guest-physical
address, it controls access to a 512-GByte region of the guest-physical-address
space.
•
A 4-KByte naturally aligned EPT page-directory-pointer table is located at the
physical address specified in bits 51:12 of the EPT PML4E (see Table 25-1). An
EPT page-directory-pointer table comprises 512 64-bit entries (PDPTEs). An EPT
PDPTE is selected using the physical address defined as follows:
Table 25-1. Format of an EPT PML4 Entry (PML4E)
Bit
Position(s)
Contents
0
Read access; indicates whether reads are allowed from the 512-GByte region
controlled by this entry
1
Write access; indicates whether writes are allowed to the 512-GByte region
controlled by this entry
2
Execute access; indicates whether instruction fetches are allowed from the 512GByte region controlled by this entry
7:3
Reserved (must be 0)
11:8
Ignored
(N–1):12
Physical address of 4-KByte aligned EPT page-directory-pointer table referenced
by this entry1
51:N
Reserved (must be 0)
63:52
Ignored
NOTES:
1. N is the physical-address width supported by the processor. Software can determine a processor’s
physical-address width by executing CPUID with 80000008H in EAX. The physical-address width
is returned in bits 7:0 of EAX.
— Bits 63:52 are all 0.
— Bits 51:12 are from the EPT PML4 entry.
— Bits 11:3 are bits 38:30 of the guest-physical address.
— Bits 2:0 are all 0.
Because a PDPTE is identified using bits 47:30 of the guest-physical address, it
controls access to a 1-GByte region of the guest-physical-address space. Use of the
PDPTE depends on the value of bit 7 in that entry:1
Vol. 3B 25-5
VMX SUPPORT FOR ADDRESS TRANSLATION
•
If bit 7 of the EPT PDPTE is 1, the EPT PDPTE maps a 1-GByte page (see
Table 25-2). The final physical address is computed as follows:
Table 25-2. Format of an EPT Page-Directory-Pointer-Table Entry (PDPTE) that Maps
a 1-GByte Page
Bit
Position(s)
Contents
0
Read access; indicates whether reads are allowed from the 1-GByte page
referenced by this entry
1
Write access; indicates whether writes are allowed to the 1-GByte page
referenced by this entry
2
Execute access; indicates whether instruction fetches are allowed from the 1GByte page referenced by this entry
5:3
EPT memory type for this 1-GByte page (see Section 25.2.4)
6
Ignore PAT memory type for this 1-GByte page (see Section 25.2.4)
7
Must be 1 (otherwise, this entry references an EPT page directory)
11:8
Ignored
29:12
Reserved (must be 0)
(N–1):30
Physical address of the 1-GByte page referenced by this entry1
51:N
Reserved (must be 0)
63:52
Ignored
NOTES:
1. N is the physical-address width supported by the logical processor.
— Bits 63:52 are all 0.
— Bits 51:30 are from the EPT PDPTE.
— Bits 29:0 are from the original guest-physical address.
•
If bit 7 of the EPT PDPTE is 0, a 4-KByte naturally aligned EPT page directory is
located at the physical address specified in bits 51:12 of the EPT PDPTE (see
Table 25-3). An EPT page-directory comprises 512 64-bit entries (PDEs). An EPT
PDE is selected using the physical address defined as follows:
1. Not all processors allow bit 7 of an EPT PDPTE to be set to 1. Software should read the VMX
capability MSR IA32_VMX_EPT_VPID_CAP (see Appendix G.10) to determine whether this is
allowed.
25-6 Vol. 3B
VMX SUPPORT FOR ADDRESS TRANSLATION
Table 25-3. Format of an EPT Page-Directory-Pointer-Table Entry (PDPTE) that
References an EPT Page Directory
Bit
Position(s)
Contents
0
Read access; indicates whether reads are allowed from the 1-GByte region
controlled by this entry
1
Write access; indicates whether writes are allowed to the 1-GByte region
controlled by this entry
2
Execute access; indicates whether instruction fetches are allowed from the 1GByte region controlled by this entry
7:3
Reserved (must be 0)
11:8
Ignored
(N–1):12
Physical address of 4-KByte aligned EPT page directory referenced by this entry1
51:N
Reserved (must be 0)
63:52
Ignored
NOTES:
1. N is the physical-address width supported by the logical processor.
— Bits 63:52 are all 0.
— Bits 51:12 are from the EPT PDPTE.
— Bits 11:3 are bits 29:21 of the guest-physical address.
— Bits 2:0 are all 0.
Because an EPT PDE is identified using bits 47:21 of the guest-physical address, it
controls access to a 2-MByte region of the guest-physical-address space. Use of the
EPT PDE depends on the value of bit 7 in that entry:
•
If bit 7 of the EPT PDE is 1, the EPT PDE maps a 2-MByte page (see Table 25-4).
The final physical address is computed as follows:
— Bits 63:52 are all 0.
— Bits 51:21 are from the EPT PDE.
— Bits 20:0 are from the original guest-physical address.
•
If bit 7 of the EPT PDE is 0, a 4-KByte naturally aligned EPT page table is located
at the physical address specified in bits 51:12 of the EPT PDE (see Table 25-5).
An EPT page table comprises 512 64-bit entries (PTEs). An EPT PTE is selected
using a physical address defined as follows:
— Bits 63:52 are all 0.
Vol. 3B 25-7
VMX SUPPORT FOR ADDRESS TRANSLATION
Table 25-4. Format of an EPT Page-Directory Entry (PDE) that Maps a 2-MByte Page
Bit
Position(s)
Contents
0
Read access; indicates whether reads are allowed from the 2-MByte page
referenced by this entry
1
Write access; indicates whether writes are allowed to the 2-MByte page
referenced by this entry
2
Execute access; indicates whether instruction fetches are allowed from the 2MByte page referenced by this entry
5:3
EPT memory type for this 2-MByte page (see Section 25.2.4)
6
Ignore PAT memory type for this 2-MByte page (see Section 25.2.4)
7
Must be 1 (otherwise, this entry references an EPT page table)
11:8
Ignored
20:12
Reserved (must be 0)
(N–1):21
Physical address of the 2-MByte page referenced by this entry1
51:N
Reserved (must be 0)
63:52
Ignored
NOTES:
1. N is the physical-address width supported by the logical processor.
— Bits 51:12 are from the EPT PDE.
— Bits 11:3 are bits 20:12 of the guest-physical address.
— Bits 2:0 are all 0.
•
Because an EPT PTE is identified using bits 47:12 of the guest-physical address,
every EPT PTE maps a 4-KByte page (see Table 25-6). The final physical address
is computed as follows:
•
•
•
Bits 63:52 are all 0.
Bits 51:12 are from the EPT PTE.
Bits 11:0 are from the original guest-physical address.
If bits 2:0 of an EPT paging-structure entry are all 0, the entry is not present. The
processor ignores bits 63:3 and does uses the entry neither to reference another EPT
paging-structure entry nor to produce a physical address. A reference using a guestphysical address whose translation encounters an EPT paging-structure that is not
present causes an EPT violation (see Section 25.2.3.2).
25-8 Vol. 3B
VMX SUPPORT FOR ADDRESS TRANSLATION
Table 25-5. Format of an EPT Page-Directory Entry (PDE) that References an EPT
Page Table
Bit
Position(s)
Contents
0
Read access; indicates whether reads are allowed from the 2-MByte region
controlled by this entry
1
Write access; indicates whether writes are allowed to the 2-MByte region
controlled by this entry
2
Execute access; indicates whether instruction fetches are allowed from the 2MByte region controlled by this entry
6:3
Reserved (must be 0)
7
Must be 0 (otherwise, this entry maps a 2-MByte page)
11:8
Ignored
(N–1):12
Physical address of 4-KByte aligned EPT page table referenced by this entry1
51:N
Reserved (must be 0)
63:52
Ignored
NOTES:
1. N is the physical-address width supported by the logical processor.
The discussion above describes how the EPT paging structures reference each other
and how the logical processor traverses those structures when translating a guestphysical address. It does not cover all details of the translation process. Additional
details are provided as follows:
•
Situations in which the translation process may lead to VM exits (sometimes
before the process completes) are described in Section 25.2.3.
•
Interactions between the EPT translation mechanism and memory typing are
described in Section 25.2.4.
Figure 25-1 gives a summary of the formats of the EPTP and the EPT paging-structure entries. For the EPT paging structure entries, it identifies separately the format
of entries that map pages, those that reference other EPT paging structures, and
those that do neither because they are “not present”; bits 2:0 and bit 7 are highlighted because they determine how a paging-structure entry is used.
Vol. 3B 25-9
VMX SUPPORT FOR ADDRESS TRANSLATION
Table 25-6. Format of an EPT Page-Table Entry
Bit
Position(s)
Contents
0
Read access; indicates whether reads are allowed from the 4-KByte page
referenced by this entry
1
Write access; indicates whether writes are allowed to the 4-KByte page
referenced by this entry
2
Execute access; indicates whether instruction fetches are allowed from the 4KByte page referenced by this entry
5:3
EPT memory type for this 4-KByte page (see Section 25.2.4)
6
Ignore PAT memory type for this 4-KByte page (see Section 25.2.4)
11:7
Ignored
(N–1):12
Physical address of the 4-KByte page referenced by this entry1
51:N
Reserved (must be 0)
63:52
Ignored
NOTES:
1. N is the physical-address width supported by the logical processor.
25.2.3
EPT-Induced VM Exits
Accesses using guest-physical addresses may cause VM exits due to EPT misconfigurations and EPT violations. An EPT misconfiguration occurs when, in the
course of translation a guest-physical address, the logical processor encounters an
EPT paging-structure entry that contains an unsupported value. An EPT violation
occurs when there is no EPT misconfiguration but the EPT paging-structure entries
disallow an access using the guest-physical address.
EPT misconfigurations and EPT violations occur only due to an attempt to access
memory with a guest-physical address. Loading CR3 with a guest-physical address
with the MOV to CR3 instruction can cause neither an EPT configuration nor an EPT
violation until that address is used to access a paging structure.1
1. If the logical processor is using PAE paging—because CR0.PG = CR4.PAE = 1 and
IA32_EFER.LMA = 0—the MOV to CR3 instruction loads the PDPTEs from memory using the
guest-physical address being loaded into CR3. In this case, therefore, the MOV to CR3 instruction
may cause an EPT misconfiguration or an EPT violation.
25-10 Vol. 3B
VMX SUPPORT FOR ADDRESS TRANSLATION
6666555555555
3210987654321
M1 M-1
Reserved
Ignored
33322222222221111111111
210987654321098765432109876543210
Address of EPT PML4 table
Rsvd.
Address of EPT page-directory-pointer table
EPT EPT
Reserved PWL– PS
1
MT
EPTP2
Ign. Reserved XW R
PML4E:
present
000
PML4E:
not
present
Ignored
Ignored
Rsvd.
Ignored
Rsvd.
Physical
address of
1GB page
Reserved
Address of EPT page directory
1
Ign.
0 Rsvd. XW R
PDPTE:
page
directory
000
PDTPE:
not
present
Ignored
Ignored
Rsvd.
Ignored
Rsvd.
Physical address
of 2MB page
Reserved
Address of EPT page table
Ign.
1
I
P EPT
XW R
A MT
T
PDE:
2MB
page
Ign.
0 Rsvd. XW R
PDE:
page
table
000
Ignored
Ignored
Rsvd.
Physical address of 4KB page
Ignored
I
PDPTE:
P EPT
XW R 1GB
A MT
page
T
Ign.
Ign.
I
P EPT XW R
A MT
T
000
PDE:
not
present
PTE:
4KB
page
PTE:
not
present
Figure 25-1. Formats of EPTP and EPT Paging-Structure Entries
NOTES:
1. M is an abbreviation for MAXPHYADDR.
2. See Section 21.6.11 for details of the EPTP.
Vol. 3B 25-11
VMX SUPPORT FOR ADDRESS TRANSLATION
25.2.3.1
EPT Misconfigurations
AN EPT misconfiguration occurs if any of the following is identified while translating a
guest-physical address:
•
The value of bits 2:0 of an EPT paging-structure entry is either 010b (write-only)
or 110b (write/execute).
•
The value of bits 2:0 of an EPT paging-structure entry is 100b (execute-only) and
this value is not supported by the logical processor. Software should read the
VMX capability MSR IA32_VMX_EPT_VPID_CAP to determine whether this value
is supported (see Appendix G.10).
•
The value of bits 2:0 of an EPT paging-structure entry is not 000b (the entry is
present) and one of the following holds:
— A reserved bit is set. This includes the setting of a bit in the range 51:12 that
is beyond the logical processor’s physical-address width.1 See Section 25.2.2
for details of which bits are reserved in which EPT paging-structure entries.
— The entry is the last one used to translate a guest physical address (either an
EPT PDE with bit 7 set to 1 or an EPT PTE) and the value of bits 5:3 (EPT
memory type) is 2, 3, or 7 (these values are reserved).
EPT misconfigurations result when an EPT paging-structure entry is configured with
settings reserved for future functionality. Software developers should be aware that
such settings may be used in the future and that an EPT paging-structure entry that
causes an EPT misconfiguration on one processor might not do so in the future.
25.2.3.2
EPT Violations
An EPT violation may occur during an access using a guest-physical address whose
translation does not cause an EPT misconfiguration. An EPT violation occurs in any of
the following situations:
•
Translation of the guest-physical address encounters an EPT paging-structure
entry that is not present (see Section 25.2.2).
•
The access is a data read and bit 0 was clear in any of the EPT paging-structure
entries used to translate the guest-physical address. Reads by the logical
processor of guest paging structures to translate a linear address are considered
to be data reads.
•
The access is a data write and bit 1 was clear in any of the EPT paging-structure
entries used to translate the guest-physical address. Writes by the logical
processor to guest paging structures to update accessed and dirty flags are
considered to be data writes.
•
The access is an instruction fetch and bit 2 was clear in any of the EPT pagingstructure entries used to translate the guest-physical address.
1. Software can determine a processor’s physical-address width by executing CPUID with
80000008H in EAX. The physical-address width is returned in bits 7:0 of EAX.
25-12 Vol. 3B
VMX SUPPORT FOR ADDRESS TRANSLATION
25.2.3.3
Prioritization of EPT-Induced VM Exits
The translation of a linear address to a physical address requires one or more translations of guest-physical addresses using EPT (see Section 25.2.1). This section
specifies the relative priority of EPT-induced VM exits with respect to each other and
to other events that may be encountered when accessing memory using a linear
address.
For an access to a guest-physical address, determination of whether an EPT misconfiguration or an EPT violation occurs is based on an iterative process:1
1. An EPT paging-structure entry is read (initially, this is an EPT PML4 entry):
a. If the entry is not present (bits 2:0 are all 0), an EPT violation occurs.
b. If the entry is present but its contents are not configured properly (see
Section 25.2.3.1), an EPT misconfiguration occurs.
c.
If the entry is present and its contents are configured properly, operation
depends on whether the entry references another EPT paging structure
(whether it is an EPT PDE with bit 7 set to 1 or an EPT PTE):
i)
If the entry does reference another EPT paging structure, an entry from
that structure is accessed; step 1 is executed for that other entry.
ii) Otherwise, the entry is used to produce the ultimate physical address
(the translation of the original guest-physical address); step 2 is
executed.
2. Once the ultimate physical address is determined, the privileges determined by
the EPT paging-structure entries are evaluated:
a. If the access to the guest-physical address is not allowed by these privileges
(see Section 25.2.3.2), an EPT violation occurs.
b. If the access to the guest-physical address is allowed by these privileges,
memory is accessed using the ultimate physical address.
If CR0.PG = 1, the translation of a linear address is also an iterative process, with the
processor first accessing an entry in the guest paging structure referenced by the
guest-physical address in CR3 (or, if PAE paging is in use, the guest-physical address
in the appropriate PDPTE register), then accessing an entry in another guest paging
structure referenced by the guest-physical address in the first guest paging-structure
entry, etc. Each guest-physical address is itself translated using EPT and may cause
an EPT-induced VM exit. The following items detail how page faults and EPT-induced
VM exits are recognized during this iterative process:
1. An attempt is made to access a guest paging-structure entry with a guestphysical address (initially, the address in CR3 or PDPTE register).
a. If the access fails because of an EPT misconfiguration or an EPT violation (see
above), an EPT-induced VM exit occurs.
1. This is a simplification of the more detailed description given in Section 25.2.2.
Vol. 3B 25-13
VMX SUPPORT FOR ADDRESS TRANSLATION
b. If the access does not cause an EPT-induced VM exit, bit 0 (the present flag)
of the entry is consulted:
i)
If the present flag is 0 or any reserved bit is set, a page fault occurs.
ii) If the present flag is 1, no reserved bit is set, operation depends on
whether the entry references another guest paging structure (whether it
is a guest PDE with PS = 1 or a guest PTE):
•
If the entry does reference another guest paging structure, an entry
from that structure is accessed; step 1 is executed for that other
entry.
•
Otherwise, the entry is used to produce the ultimate guest-physical
address (the translation of the original linear address); step 2 is
executed.
2. Once the ultimate guest-physical address is determined, the privileges
determined by the guest paging-structure entries are evaluated:
a. If the access to the linear address is not allowed by these privileges (e.g., it
was a write to a read-only page), a page fault occurs.
b. If the access to the linear address is allowed by these privileges, an attempt
is made to access memory at the ultimate guest-physical address:
i)
If the access fails because of an EPT misconfiguration or an EPT violation
(see above), an EPT-induced VM exit occurs.
ii) If the access does not cause an EPT-induced VM exit, memory is accessed
using the ultimate physical address (the translation, using EPT, of the
ultimate guest-physical address).
If CR0.PG = 0, a linear address is treated as a guest-physical address and is translated using EPT (see above). This process, if it completes without an EPT violation or
EPT misconfiguration, produces a physical address and determines the privileges
allowed by the EPT paging-structure entries. If these privileges do not allow the
access to the physical address (see Section 25.2.3.2), an EPT violation occurs.
Otherwise, memory is accessed using the physical address.
25.2.4
EPT and Memory Typing
This section specifies how a logical processor determines the memory type use for a
memory access while EPT is in use. (See Chapter 11, “Memory Cache Control” of
Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 3A for
details of memory typing in the Intel 64 architecture.) Section 25.2.4.1 explains how
the memory type is determined for accesses to the EPT paging structures. Section
25.2.4.2 explains how the memory type is determined for an access using a guestphysical address that is translated using EPT.
25-14 Vol. 3B
VMX SUPPORT FOR ADDRESS TRANSLATION
25.2.4.1
Memory Type Used for Accessing EPT Paging Structures
This section explains how the memory type is determined for accesses to the EPT
paging structures. The determination is based first on the value of bit 30 (cache
disable—CD) in control register CR0:
•
If CR0.CD = 0, the memory type used for any such reference is the EPT pagingstructure memory type, which is specified in bits 2:0 of the extended-page-table
pointer (EPTP), a VM-execution control field (see Section 21.6.11). A value of 0
indicates the uncacheable type (UC), while a value of 6 indicates the write-back
type (WB). Other values are reserved.
•
If CR0.CD = 1, the memory type used for any such reference is uncacheable
(UC).
The MTRRs have no effect on the memory type used for an access to an EPT paging
structure.
25.2.4.2
Memory Type Used for Translated Guest-Physical Addresses
The effective memory type of a memory access using a guest-physical address (an
access that is translated using EPT) is the memory type that is used to access
memory. The effective memory type is based on the value of bit 30 (cache
disable—CD) in control register CR0; the last EPT paging-structure entry used to
translate the guest-physical address (either an EPT PDE with bit 7 set to 1 or an EPT
PTE); and the PAT memory type (see below):
•
The PAT memory type depends on the value of CR0.PG:
— If CR0.PG = 0, the PAT memory type is WB (writeback).1
— If CR0.PG = 1, the PAT memory type is the memory type selected from the
IA32_PAT MSR as specified in Section 11.12.3, “Selecting a Memory Type
from the PAT”.2
•
The EPT memory type is specified in bits 5:3 of the last EPT paging-structure
entry: 0 = UC; 1 = WC; 4 = WT; 5 = WP; and 6 = WB. Other values are reserved
and cause EPT misconfigurations (see Section 25.2.3).
•
If CR0.CD = 0, the effective memory type depends upon the value of bit 6 of the
last EPT paging-structure entry:
1. If the capability MSR IA32_VMX_CR0_FIXED0 reports that CR0.PG must be 1 in VMX operation,
CR0.PG can be 0 in VMX non-root operation only if the “unrestricted guest” VM-execution control
and bit 31 of the primary processor-based VM-execution controls are both 1.
2. Table 11-11 in Section 11.12.3, “Selecting a Memory Type from the PAT” illustrates how the PAT
memory type is selected based on the values of the PAT, PCD, and PWT bits in a page-table entry
(or page-directory entry with PS = 1). For accesses to a guest paging-structure entry X, the PAT
memory type is selected from the table by using a value of 0 for the PAT bit with the values of
PCD and PWT from the paging-structure entry Y that references X (or from CR3 if X is in the root
paging structure). With PAE paging, the PAT memory type for accesses to the PDPTEs is WB.
Vol. 3B 25-15
VMX SUPPORT FOR ADDRESS TRANSLATION
— If the value is 0, the effective memory type is the combination of the EPT
memory type and the PAT memory type specified in Table 11-7 in Section
11.5.2.2, using the EPT memory type in place of the MTRR memory type.
— If the value is 1, the memory type used for the access is the EPT memory
type. The PAT memory type is ignored.
•
If CR0.CD = 1, the effective memory type is UC.
The MTRRs have no effect on the memory type used for an access to a guest-physical
address.
25.3
CACHING TRANSLATION INFORMATION
Processors supporting Intel® 64 and IA-32 architectures may accelerate the
address-translation process by caching on the processor data from the structures in
memory that control that process. Such caching is discussed in Section 4.10,
“Caching Translation Information” in the Intel® 64 and IA-32 Architectures Software
Developer’s Manual, Volume 3A. The current section describes how this caching
interacts with the VMX architecture.
The VPID and EPT features of the architecture for VMX operation augment this
caching architecture. EPT defines the guest-physical address space and defines
translations to that address space (from the linear-address space) and from that
address space (to the physical-address space). Both features control the ways in
which a logical processor may create and use information cached from the paging
structures.
Section 25.3.1 describes the different kinds of information that may be cached.
Section 25.3.2 specifies when such information may be cached and how it may be
used. Section 25.3.3 details how software can invalidate cached information.
25.3.1
Information That May Be Cached
Section 4.10, “Caching Translation Information” in Intel® 64 and IA-32 Architectures
Software Developer’s Manual, Volume 3A identifies two kinds of translation-related
information that may be cached by a logical processor: translations, which are
mappings from linear page numbers to physical page frames, and paging-structure
caches, which map the upper bits of a linear page number to information from the
paging-structure entries used to translate linear addresses matching those upper
bits.
The same kinds of information may be cached when VPIDs and EPT are in use. A
logical processor may cache and use such information based on its function. Information with different functionality is identified as follows:
•
Linear mappings.1 There are two kinds:
1. Earlier versions of this manual used the term “VPID-tagged” to identify linear mappings.
25-16 Vol. 3B
VMX SUPPORT FOR ADDRESS TRANSLATION
— Linear translations. Each of these is a mapping from a linear page number to
the physical page frame to which it translates, along with information about
access privileges and memory typing.
— Linear paging-structure-cache entries. Each of these is a mapping from the
upper portion of a linear address to the physical address of the paging
structure used to translate the corresponding region of the linear-address
space, along with information about access privileges. For example,
bits 47:39 of a linear address would map to the address of the relevant pagedirectory-pointer table.
Linear mappings do not contain information from any EPT paging structure.
•
Guest-physical mappings.1 There are two kinds:
— Guest-physical translations. Each of these is a mapping from a guest-physical
page number to the physical page frame to which it translates, along with
information about access privileges and memory typing.
— Guest-physical paging-structure-cache entries. Each of these is a mapping
from the upper portion of a guest-physical address to the physical address of
the EPT paging structure used to translate the corresponding region of the
guest-physical address space, along with information about access
privileges.
The information in guest-physical mappings about access privileges and memory
typing is derived from EPT paging structures.
•
Combined mappings.2 There are two kinds:
— Combined translations. Each of these is a mapping from a linear page number
to the physical page frame to which it translates, along with information
about access privileges and memory typing.
— Combined paging-structure-cache entries. Each of these is a mapping from
the upper portion of a linear address to the physical address of the paging
structure used to translate the corresponding region of the linear-address
space, along with information about access privileges.
The information in combined mappings about access privileges and memory
typing is derived from both guest paging structures and EPT paging structures.
25.3.2
Creating and Using Cached Translation Information
The following items detail the creation of the mappings described in the previous
section:3
•
The following items describe the creation of mappings while EPT is not in use
(including execution outside VMX non-root operation):
1. Earlier versions of this manual used the term “EPTP-tagged” to identify guest-physical mappings.
2. Earlier versions of this manual used the term “dual-tagged” to identify combined mappings.
Vol. 3B 25-17
VMX SUPPORT FOR ADDRESS TRANSLATION
— Linear mappings may be created. They are derived from the paging
structures referenced (directly or indirectly) by the current value of CR3 and
are associated with the current VPID and the current PCID.
— No linear mappings are created with information derived from pagingstructure entries that are not present (bit 0 is 0) or that set reserved bits. For
example, if a PTE is not present, no linear mapping are created for any linear
page number whose translation would use that PTE.
— No guest-physical or combined mappings are created while EPT is not in use.
•
The following items describe the creation of mappings while EPT is in use:
— Guest-physical mappings may be created. They are derived from the EPT
paging structures referenced (directly or indirectly) by bits 51:12 of the
current EPTP. These 40 bits contain the address of the EPT-PML4-table. (the
notation EP4TA refers to those 40 bits). Newly created guest-physical
mappings are associated with the current EP4TA.
— Combined mappings may be created. They are derived from the EPT paging
structures referenced (directly or indirectly) by the current EP4TA. If
CR0.PG = 1, they are also derived from the paging structures referenced
(directly or indirectly) by the current value of CR3. They are associated with
the current VPID, the current PCID, and the current EP4TA.1 No combined
paging-structure-cache entries are created if CR0.PG = 0.2
— No guest-physical mappings or combined mappings are created with
information derived from EPT paging-structure entries that are not present
(bits 2:0 are all 0) or that are misconfigured (see Section 25.2.3.1).
— No combined mappings are created with information derived from guest
paging-structure entries that are not present or that set reserved bits.
— No linear mappings are created while EPT is in use.
The following items detail the use of the various mappings:
•
If EPT is not in use (e.g., when outside VMX non-root operation), a logical
processor may use cached mappings as follows:
3. This section associated cached information with the current VPID and PCID. If PCIDs are not supported or are not being used (e.g., because CR4.PCIDE = 0), all the information is implicitly associated with PCID 000H; see Section 4.10.1, “Process-Context Identifiers (PCIDs),” in Intel® 64 and
IA-32 Architectures Software Developer’s Manual, Volume 3A.
1. At any given time, a logical processor may be caching combined mappings for a VPID and a PCID
that are associated with different EP4TAs. Similarly, it may be caching combined mappings for an
EP4TA that are associated with different VPIDs and PCIDs.
2. If the capability MSR IA32_VMX_CR0_FIXED0 reports that CR0.PG must be 1 in VMX operation,
CR0.PG can be 0 in VMX non-root operation only if the “unrestricted guest” VM-execution control
and bit 31 of the primary processor-based VM-execution controls are both 1.
25-18 Vol. 3B
VMX SUPPORT FOR ADDRESS TRANSLATION
— For accesses using linear addresses, it may use linear mappings associated
with the current VPID and the current PCID. It may also use global TLB
entries (linear mappings) associated with the current VPID and any PCID.
— No guest-physical or combined mappings are used while EPT is not in use.
•
If EPT is in use, a logical processor may use cached mappings as follows:
— For accesses using linear addresses, it may use combined mappings
associated with the current VPID, the current PCID, and the current EP4TA. It
may also use global TLB entries (combined mappings) associated with the
current VPID, the current EP4TA, and any PCID.
— For accesses using guest-physical addresses, it may use guest-physical
mappings associated with the current EP4TA.
— No linear mappings are used while EPT is in use.
25.3.3
Invalidating Cached Translation Information
Software modifications of paging structures (including EPT paging structures) may
result in inconsistencies between those structures and the mappings cached by a
logical processor. Certain operations invalidate information cached by a logical
processor and can be used to eliminate such inconsistencies.
25.3.3.1
Operations that Invalidate Cached Mappings
The following operations invalidate cached mappings as indicated:
•
Operations that architecturally invalidate entries in the TLBs or paging-structure
caches independent of VMX operation (e.g., the INVLPG instruction) invalidate
linear mappings and combined mappings.1 They are required to do so only for the
current VPID (but, for combined mappings, all EP4TAs). Linear mappings for the
current VPID are invalidated even if EPT is in use.2 Combined mappings for the
current VPID are invalidated even if EPT is not in use.3
•
An EPT violation invalidates any guest-physical mappings (associated with the
current EP4TA) that would be used to translate the guest-physical address that
1. See Section 4.10.4, “Invalidation of TLBs and Paging-Structure Caches,” in the Intel® 64 and IA-32
Architectures Software Developer’s Manual, Volume 3A for an enumeration of operations that
architecturally invalidate entries in the TLBs and paging-structure caches independent of VMX
operation.
2. While no linear mappings are created while EPT is in use, a logical processor may retain, while
EPT is in use, linear mappings (for the same VPID as the current one) there were created earlier,
when EPT was not in use.
3. While no combined mappings are created while EPT is not in use, a logical processor may retain,
while EPT is in not use, combined mappings (for the same VPID as the current one) there were
created earlier, when EPT was in use.
Vol. 3B 25-19
VMX SUPPORT FOR ADDRESS TRANSLATION
caused the EPT violation. If that guest-physical address was the translation of a
linear address, the EPT violation also invalidates any combined mappings for that
linear address associated with the current PCID, the current VPID and the current
EP4TA.
•
If the “enable VPID” VM-execution control is 0, VM entries and VM exits
invalidate linear mappings and combined mappings associated with VPID 0000H
(for all PCIDs). Combined mappings for VPID 0000H are invalidated for all
EP4TAs.
•
Execution of the INVVPID instruction invalidates linear mappings and combined
mappings. Invalidation is based on instruction operands, called the INVVPID type
and the INVVPID descriptor. Four INVVPID types are currently defined:
— Individual-address. If the INVVPID type is 0, the logical processor
invalidates linear mappings and combined mappings associated with the
VPID specified in the INVVPID descriptor and that would be used to translate
the linear address specified in of the INVVPID descriptor. Linear mappings
and combined mappings for that VPID and linear address are invalidated for
all PCIDs and, for combined mappings, all EP4TAs. (The instruction may also
invalidate mappings associated with other VPIDs and for other linear
addresses.)
— Single-context. If the INVVPID type is 1, the logical processor invalidates all
linear mappings and combined mappings associated with the VPID specified
in the INVVPID descriptor. Linear mappings and combined mappings for that
VPID are invalidated for all PCIDs and, for combined mappings, all EP4TAs.
(The instruction may also invalidate mappings associated with other VPIDs.)
— All-context. If the INVVPID type is 2, the logical processor invalidates linear
mappings and combined mappings associated with all VPIDs except VPID
0000H and with all PCIDs. (In some cases, it may invalidate linear mappings
with VPID 0000H as well.) Combined mappings are invalidated for all EP4TAs.
— Single-context-retaining-globals. If the INVVPID type is 3, the logical
processor invalidates linear mappings and combined mappings associated
with the VPID specified in the INVVPID descriptor. Linear mappings and
combined mappings for that VPID are invalidated for all PCIDs and, for
combined mappings, all EP4TAs. The logical processor is not required to
invalidate information that was used for global translations (although it may
do so). See Section 4.10, “Caching Translation Information” for details
regarding global translations. (The instruction may invalidate mappings
associated with other VPIDs.)
See Chapter 5 of the Intel® 64 and IA-32 Architectures Software Developer’s
Manual, Volume 2B for details of the INVVPID instruction. See Section 25.3.3.3
for guidelines regarding use of this instruction.
•
Execution of the INVEPT instruction invalidates guest-physical mappings and
combined mappings. Invalidation is based on instruction operands, called the
INVEPT type and the INVEPT descriptor. Two INVEPT types are currently defined:
25-20 Vol. 3B
VMX SUPPORT FOR ADDRESS TRANSLATION
— Single-context. If the INVEPT type is 1, the logical processor invalidates all
guest-physical mappings and combined mappings associated with the EP4TA
specified in the INVEPT descriptor. Combined mappings for that EP4TA are
invalidated for all VPIDs and all PCIDs. (The instruction may invalidate
mappings associated with other EP4TAs.)
— All-context. If the INVEPT type is 2, the logical processor invalidates guestphysical mappings and combined mappings associated with all EP4TAs (and,
for combined mappings, for all VPIDs and PCIDs).
See Chapter 5 of the Intel® 64 and IA-32 Architectures Software Developer’s
Manual, Volume 2B for details of the INVEPT instruction. See Section 25.3.3.4 for
guidelines regarding use of this instruction.
•
A power-up or a reset invalidates all linear mappings, guest-physical mappings,
and combined mappings.
25.3.3.2
Operations that Need Not Invalidate Cached Mappings
The following items detail cases of operations that are not required to invalidate
certain cached mappings:
•
Operations that architecturally invalidate entries in the TLBs or paging-structure
caches independent of VMX operation are not required to invalidate any guestphysical mappings.
•
The INVVPID instruction is not required to invalidate any guest-physical
mappings.
•
•
The INVEPT instruction is not required to invalidate any linear mappings.
•
The VMXOFF and VMXON instructions are not required to invalidate any linear
mappings, guest-physical mappings, or combined mappings.
VMX transitions are not required to invalidate any guest-physical mappings. If
the “enable VPID” VM-execution control is 1, VMX transitions are not required to
invalidate any linear mappings or combined mappings.
A logical processor may invalidate any cached mappings at any time. For this reason,
the operations identified above may invalidate the indicated mappings despite the
fact that doing so is not required.
25.3.3.3
Guidelines for Use of the INVVPID Instruction
The need for VMM software to use the INVVPID instruction depends on how that software is virtualizing memory (e.g., see Section 28.3, “Memory Virtualization”).
If EPT is not in use, it is likely that the VMM is virtualizing the guest paging structures.
Such a VMM may configure the VMCS so that all or some of the operations that invalidate entries the TLBs and the paging-structure caches (e.g., the INVLPG instruction)
cause VM exits. If VMM software is emulating these operations, it may be necessary
to use the INVVPID instruction to ensure that the logical processor’s TLBs and the
paging-structure caches are appropriately invalidated.
Vol. 3B 25-21
VMX SUPPORT FOR ADDRESS TRANSLATION
Requirements of when software should use the INVVPID instruction depend on the
specific algorithm being used for page-table virtualization. The following items
provide guidelines for software developers:
•
Emulation of the INVLPG instruction may require execution of the INVVPID
instruction as follows:
— The INVVPID type is individual-address (0).
— The VPID in the INVVPID descriptor is the one assigned to the virtual
processor whose execution is being emulated.
— The linear address in the INVVPID descriptor is that of the operand of the
INVLPG instruction being emulated.
•
Some instructions invalidate all entries in the TLBs and paging-structure
caches—except for global translations. An example is the MOV to CR3 instruction.
(See Section 4.10, “Caching Translation Information” in the Intel® 64 and IA-32
Architectures Software Developer’s Manual, Volume 3A for details regarding
global translations.) Emulation of such an instruction may require execution of
the INVVPID instruction as follows:
— The INVVPID type is single-context-retaining-globals (3).
— The VPID in the INVVPID descriptor is the one assigned to the virtual
processor whose execution is being emulated.
•
Some instructions invalidate all entries in the TLBs and paging-structure
caches—including for global translations. An example is the MOV to CR4
instruction if the value of value of bit 4 (page global enable—PGE) is changing.
Emulation of such an instruction may require execution of the INVVPID
instruction as follows:
— The INVVPID type is single-context (1).
— The VPID in the INVVPID descriptor is the one assigned to the virtual
processor whose execution is being emulated.
If EPT is not in use, the logical processor associates all mappings it creates with the
current VPID, and it will use such mappings to translate linear addresses. For that
reason, a VMM should not use the same VPID for different non-EPT guests that use
different page tables. Doing so may result in one guest using translations that pertain
to the other.
If EPT is in use, the instructions enumerated above might not be configured to cause
VM exits and the VMM might not be emulating them. In that case, executions of the
instructions by guest software properly invalidate the required entries in the TLBs
and paging-structure caches (see Section 25.3.3.1); execution of the INVVPID
instruction is not required.
If EPT is in use, the logical processor associates all mappings it creates with the value
of bits 51:12 of current EPTP. If a VMM uses different EPTP values for different guests,
it may use the same VPID for those guests. Doing so cannot result in one guest using
translations that pertain to the other.
25-22 Vol. 3B
VMX SUPPORT FOR ADDRESS TRANSLATION
The following guidelines apply more generally and are appropriate even if EPT is in
use:
•
As detailed in Section 22.2.1.1, an access to the APIC-access page might not
cause an APIC-access VM exit if software does not properly invalidate information
that may be cached from the paging structures. If, at one time, the current VPID
on a logical processor was a non-zero value X, it is recommended that software
use the INVVPID instruction with the “single-context” INVVPID type and with
VPID X in the INVVPID descriptor before a VM entry on the same logical
processor that establishes VPID X and either (a) the “virtualize APIC accesses”
VM-execution control was changed from 0 to 1; or (b) the value of the APICaccess address was changed.
•
Software can use the INVVPID instruction with the “all-context” INVVPID type
immediately after execution of the VMXON instruction or immediately prior to
execution of the VMXOFF instruction. Either prevents potentially undesired
retention of information cached from paging structures between separate uses of
VMX operation.
25.3.3.4
Guidelines for Use of the INVEPT Instruction
The following items provide guidelines for use of the INVEPT instruction to invalidate
information cached from the EPT paging structures.
•
Software should use the INVEPT instruction with the “single-context” INVEPT
type after making any of the following changes to an EPT paging-structure entry
(the INVEPT descriptor should contain an EPTP value that references — directly
or indirectly — the modified EPT paging structure):
— Changing any of the privilege bits 2:0 from 1 to 0.
— Changing the physical address in bits 51:12.
— For an EPT PDPTE or an EPT PDE, changing bit 7 (which determines whether
the entry maps a page).
— For the last EPT paging-structure entry used to translate a guest-physical
address (either an EPT PDE with bit 7 set to 1 or an EPT PTE), changing either
bits 5:3 or bit 6. (These bits determine the effective memory type of
accesses using that EPT paging-structure entry; see Section 25.2.4.)
•
Software may use the INVEPT instruction after modifying a present EPT pagingstructure entry to change any of the privilege bits 2:0 from 0 to 1. Failure to do
so may cause an EPT violation that would not otherwise occur. Because an EPT
violation invalidates any mappings that would be used by the access that caused
the EPT violation (see Section 25.3.3.1), an EPT violation will not recur if the
original access is performed again, even if the INVEPT instruction is not executed.
•
Because a logical processor does not cache any information derived from EPT
paging-structure entries that are not present or misconfigured (see Section
25.2.3.1), it is not necessary to execute INVEPT following modification of an EPT
paging-structure entry that had been not present or misconfigured.
Vol. 3B 25-23
VMX SUPPORT FOR ADDRESS TRANSLATION
•
As detailed in Section 22.2.1.1 and Section 22.2.2.1, an access to the APICaccess page might not cause an APIC-access VM exit if software does not
properly invalidate information that may be cached from the EPT paging
structures. If EPT was in use on a logical processor at one time with EPTP X, it is
recommended that software use the INVEPT instruction with the “single-context”
INVEPT type and with EPTP X in the INVEPT descriptor before a VM entry on the
same logical processor that enables EPT with EPTP X and either (a) the “virtualize
APIC accesses” VM-execution control was changed from 0 to 1; or (b) the value
of the APIC-access address was changed.
•
Software can use the INVEPT instruction with the “all-context” INVEPT type
immediately after execution of the VMXON instruction or immediately prior to
execution of the VMXOFF instruction. Either prevents potentially undesired
retention of information cached from EPT paging structures between separate
uses of VMX operation.
In a system containing more than one logical processor, software must account for
the fact that information from an EPT paging-structure entry may be cached on
logical processors other than the one that modifies that entry. The process of propagating the changes to a paging-structure entry is commonly referred to as “TLB
shootdown.” A discussion of TLB shootdown appears in Section 4.10.5, “Propagation
of Paging-Structure Changes to Multiple Processors,” in the Intel® 64 and IA-32
Architectures Software Developer’s Manual, Volume 3A.
25-24 Vol. 3B
CHAPTER 26
SYSTEM MANAGEMENT MODE
This chapter describes aspects of IA-64 and IA-32 architecture used in system
management mode (SMM).
SMM provides an alternate operating environment that can be used to monitor and
manage various system resources for more efficient energy usage, to control system
hardware, and/or to run proprietary code. It was introduced into the IA-32 architecture in the Intel386 SL processor (a mobile specialized version of the Intel386
processor). It is also available in the Pentium M, Pentium 4, Intel Xeon, P6 family, and
Pentium and Intel486 processors (beginning with the enhanced versions of the
Intel486 SL and Intel486 processors).
26.1
SYSTEM MANAGEMENT MODE OVERVIEW
SMM is a special-purpose operating mode provided for handling system-wide functions like power management, system hardware control, or proprietary OEMdesigned code. It is intended for use only by system firmware, not by applications
software or general-purpose systems software. The main benefit of SMM is that it
offers a distinct and easily isolated processor environment that operates transparently to the operating system or executive and software applications.
When SMM is invoked through a system management interrupt (SMI), the processor
saves the current state of the processor (the processor’s context), then switches to a
separate operating environment contained in system management RAM (SMRAM).
While in SMM, the processor executes SMI handler code to perform operations such
as powering down unused disk drives or monitors, executing proprietary code, or
placing the whole system in a suspended state. When the SMI handler has completed
its operations, it executes a resume (RSM) instruction. This instruction causes the
processor to reload the saved context of the processor, switch back to protected or
real mode, and resume executing the interrupted application or operating-system
program or task.
The following SMM mechanisms make it transparent to applications programs and
operating systems:
•
•
The only way to enter SMM is by means of an SMI.
•
Upon entering SMM, the processor saves the context of the interrupted program
or task.
The processor executes SMM code in a separate address space (SMRAM) that can
be made inaccessible from the other operating modes.
Vol. 3B 26-1
SYSTEM MANAGEMENT MODE
•
All interrupts normally handled by the operating system are disabled upon entry
into SMM.
•
The RSM instruction can be executed only in SMM.
SMM is similar to real-address mode in that there are no privilege levels or address
mapping. An SMM program can address up to 4 GBytes of memory and can execute
all I/O and applicable system instructions. See Section 26.5 for more information
about the SMM execution environment.
NOTES
Software developers should be aware that, even if a logical processor
was using the physical-address extension (PAE) mechanism
(introduced in the P6 family processors) or was in IA-32e mode
before an SMI, this will not be the case after the SMI is delivered. This
is because delivery of an SMI disables paging (see Table 26-4). (This
does not apply if the dual-monitor treatment of SMIs and SMM is
active; see Section 26.15.)
26.1.1
System Management Mode and VMX Operation
Traditionally, SMM services system management interrupts and then resumes
program execution (back to the software stack consisting of executive and application software; see Section 26.2 through Section 26.13).
A virtual machine monitor (VMM) using VMX can act as a host to multiple virtual
machines and each virtual machine can support its own software stack of executive
and application software. On processors that support VMX, virtual-machine extensions may use system-management interrupts (SMIs) and system-management
mode (SMM) in one of two ways:
•
Default treatment. System firmware handles SMIs. The processor saves architectural states and critical states relevant to VMX operation upon entering SMM.
When the firmware completes servicing SMIs, it uses RSM to resume VMX
operation.
•
Dual-monitor treatment. Two VM monitors collaborate to control the servicing
of SMIs: one VMM operates outside of SMM to provide basic virtualization in
support for guests; the other VMM operates inside SMM (while in VMX operation)
to support system-management functions. The former is referred to as
executive monitor, the latter SMM monitor.1
The default treatment is described in Section 26.14, “Default Treatment of SMIs and
SMM with VMX Operation and SMX Operation”. Dual-monitor treatment of SMM is
described in Section 26.15, “Dual-Monitor Treatment of SMIs and SMM”.
1. The dual-monitor treatment may not be supported by all processors. Software should consult the
VMX capability MSR IA32_VMX_BASIC (see Appendix G.1) to determine whether it is supported.
26-2 Vol. 3B
SYSTEM MANAGEMENT MODE
26.2
SYSTEM MANAGEMENT INTERRUPT (SMI)
The only way to enter SMM is by signaling an SMI through the SMI# pin on the
processor or through an SMI message received through the APIC bus. The SMI is a
nonmaskable external interrupt that operates independently from the processor’s
interrupt- and exception-handling mechanism and the local APIC. The SMI takes
precedence over an NMI and a maskable interrupt. SMM is non-reentrant; that is, the
SMI is disabled while the processor is in SMM.
NOTES
In the Pentium 4, Intel Xeon, and P6 family processors, when a
processor that is designated as an application processor during an MP
initialization sequence is waiting for a startup IPI (SIPI), it is in a
mode where SMIs are masked. However if a SMI is received while an
application processor is in the wait for SIPI mode, the SMI will be
pended. The processor then responds on receipt of a SIPI by
immediately servicing the pended SMI and going into SMM before
handling the SIPI.
An SMI may be blocked for one instruction following execution of STI,
MOV to SS, or POP into SS.
26.3
SWITCHING BETWEEN SMM AND THE OTHER
PROCESSOR OPERATING MODES
Figure 2-3 shows how the processor moves between SMM and the other processor
operating modes (protected, real-address, and virtual-8086). Signaling an SMI while
the processor is in real-address, protected, or virtual-8086 modes always causes the
processor to switch to SMM. Upon execution of the RSM instruction, the processor
always returns to the mode it was in when the SMI occurred.
26.3.1
Entering SMM
The processor always handles an SMI on an architecturally defined “interruptible”
point in program execution (which is commonly at an IA-32 architecture instruction
boundary). When the processor receives an SMI, it waits for all instructions to retire
and for all stores to complete. The processor then saves its current context in SMRAM
(see Section 26.4), enters SMM, and begins to execute the SMI handler.
Upon entering SMM, the processor signals external hardware that SMM handling has
begun. The signaling mechanism used is implementation dependent. For the P6
family processors, an SMI acknowledge transaction is generated on the system bus
and the multiplexed status signal EXF4 is asserted each time a bus transaction is
generated while the processor is in SMM. For the Pentium and Intel486 processors,
the SMIACT# pin is asserted.
Vol. 3B 26-3
SYSTEM MANAGEMENT MODE
An SMI has a greater priority than debug exceptions and external interrupts. Thus, if
an NMI, maskable hardware interrupt, or a debug exception occurs at an instruction
boundary along with an SMI, only the SMI is handled. Subsequent SMI requests are
not acknowledged while the processor is in SMM. The first SMI interrupt request that
occurs while the processor is in SMM (that is, after SMM has been acknowledged to
external hardware) is latched and serviced when the processor exits SMM with the
RSM instruction. The processor will latch only one SMI while in SMM.
See Section 26.5 for a detailed description of the execution environment when in
SMM.
26.3.2
Exiting From SMM
The only way to exit SMM is to execute the RSM instruction. The RSM instruction is
only available to the SMI handler; if the processor is not in SMM, attempts to execute
the RSM instruction result in an invalid-opcode exception (#UD) being generated.
The RSM instruction restores the processor’s context by loading the state save image
from SMRAM back into the processor’s registers. The processor then returns an
SMIACK transaction on the system bus and returns program control back to the
interrupted program.
Upon successful completion of the RSM instruction, the processor signals external
hardware that SMM has been exited. For the P6 family processors, an SMI acknowledge transaction is generated on the system bus and the multiplexed status signal
EXF4 is no longer generated on bus cycles. For the Pentium and Intel486 processors,
the SMIACT# pin is deserted.
If the processor detects invalid state information saved in the SMRAM, it enters the
shutdown state and generates a special bus cycle to indicate it has entered shutdown
state. Shutdown happens only in the following situations:
•
A reserved bit in control register CR4 is set to 1 on a write to CR4. This error
should not happen unless SMI handler code modifies reserved areas of the
SMRAM saved state map (see Section 26.4.1). CR4 is saved in the state map in a
reserved location and cannot be read or modified in its saved state.
•
An illegal combination of bits is written to control register CR0, in particular PG
set to 1 and PE set to 0, or NW set to 1 and CD set to 0.
•
•
CR4.PCIDE would be set to 1 and IA32_EFER.LMA to 0.
(For the Pentium and Intel486 processors only.) If the address stored in the
SMBASE register when an RSM instruction is executed is not aligned on a
32-KByte boundary. This restriction does not apply to the P6 family processors.
In the shutdown state, Intel processors stop executing instructions until a RESET#,
INIT# or NMI# is asserted. While Pentium family processors recognize the SMI#
signal in shutdown state, P6 family and Intel486 processors do not. Intel does not
support using SMI# to recover from shutdown states for any processor family; the
response of processors in this circumstance is not well defined. On Pentium 4 and
later processors, shutdown will inhibit INTR and A20M but will not change any of the
26-4 Vol. 3B
SYSTEM MANAGEMENT MODE
other inhibits. On these processors, NMIs will be inhibited if no action is taken in the
SMM handler to uninhibit them (see Section 26.8).
If the processor is in the HALT state when the SMI is received, the processor handles
the return from SMM slightly differently (see Section 26.10). Also, the SMBASE
address can be changed on a return from SMM (see Section 26.11).
26.4
SMRAM
While in SMM, the processor executes code and stores data in the SMRAM space. The
SMRAM space is mapped to the physical address space of the processor and can be
up to 4 GBytes in size. The processor uses this space to save the context of the
processor and to store the SMI handler code, data and stack. It can also be used to
store system management information (such as the system configuration and
specific information about powered-down devices) and OEM-specific information.
The default SMRAM size is 64 KBytes beginning at a base physical address in physical
memory called the SMBASE (see Figure 26-1). The SMBASE default value following a
hardware reset is 30000H. The processor looks for the first instruction of the SMI
handler at the address [SMBASE + 8000H]. It stores the processor’s state in the area
from [SMBASE + FE00H] to [SMBASE + FFFFH]. See Section 26.4.1 for a description
of the mapping of the state save area.
The system logic is minimally required to decode the physical address range for the
SMRAM from [SMBASE + 8000H] to [SMBASE + FFFFH]. A larger area can be
decoded if needed. The size of this SMRAM can be between 32 KBytes and 4 GBytes.
The location of the SMRAM can be changed by changing the SMBASE value (see
Section 26.11). It should be noted that all processors in a multiple-processor system
are initialized with the same SMBASE value (30000H). Initialization software must
sequentially place each processor in SMM and change its SMBASE so that it does not
overlap those of other processors.
The actual physical location of the SMRAM can be in system memory or in a separate
RAM memory. The processor generates an SMI acknowledge transaction (P6 family
processors) or asserts the SMIACT# pin (Pentium and Intel486 processors) when the
processor receives an SMI (see Section 26.3.1).
System logic can use the SMI acknowledge transaction or the assertion of the
SMIACT# pin to decode accesses to the SMRAM and redirect them (if desired) to
specific SMRAM memory. If a separate RAM memory is used for SMRAM, system logic
should provide a programmable method of mapping the SMRAM into system memory
space when the processor is not in SMM. This mechanism will enable start-up procedures to initialize the SMRAM space (that is, load the SMI handler) before executing
the SMI handler during SMM.
Vol. 3B 26-5
SYSTEM MANAGEMENT MODE
26.4.1
SMRAM State Save Map
When an IA-32 processor that does not support Intel 64 architecture initially enters
SMM, it writes its state to the state save area of the SMRAM. The state save area
begins at [SMBASE + 8000H + 7FFFH] and extends down to [SMBASE + 8000H +
7E00H]. Table 26-1 shows the state save map. The offset in column 1 is relative to
the SMBASE value plus 8000H. Reserved spaces should not be used by software.
Some of the registers in the SMRAM state save area (marked YES in column 3) may
be read and changed by the SMI handler, with the changed values restored to the
processor registers by the RSM instruction. Some register images are read-only, and
must not be modified (modifying these registers will result in unpredictable
behavior). An SMI handler should not rely on any values stored in an area that is
marked as reserved.
SMRAM
SMBASE + FFFFH
SMBASE + 8000H
Start of State Save Area
SMI Handler Entry Point
SMBASE
Figure 26-1. SMRAM Usage
Table 26-1. SMRAM State Save Map
Offset
(Added to SMBASE +
8000H)
Register
Writable?
7FFCH
CR0
No
7FF8H
CR3
No
7FF4H
EFLAGS
Yes
7FF0H
EIP
Yes
7FECH
EDI
Yes
7FE8H
ESI
Yes
7FE4H
EBP
Yes
7FE0H
ESP
Yes
26-6 Vol. 3B
SYSTEM MANAGEMENT MODE
Table 26-1. SMRAM State Save Map (Contd.)
Offset
(Added to SMBASE +
8000H)
Register
Writable?
7FDCH
EBX
Yes
7FD8H
EDX
Yes
7FD4H
ECX
Yes
7FD0H
EAX
Yes
7FCCH
DR6
No
7FC8H
DR7
No
7FC4H
1
TR
No
7FC0H
Reserved
No
7FBCH
GS1
No
7FB8H
1
No
7FB4H
DS
1
No
7FB0H
SS1
No
7FACH
1
No
1
FS
CS
7FA8H
ES
No
7FA4H
I/O State Field, see Section 26.7
No
7FA0H
I/O Memory Address Field, see Section 26.7
No
7F9FH-7F03H
Reserved
No
7F02H
Auto HALT Restart Field (Word)
Yes
7F00H
I/O Instruction Restart Field (Word)
Yes
7EFCH
SMM Revision Identifier Field (Doubleword)
No
7EF8H
SMBASE Field (Doubleword)
Yes
7EF7H - 7E00H
Reserved
No
NOTE:
1. The two most significant bytes are reserved.
The following registers are saved (but not readable) and restored upon exiting SMM:
•
•
Control register CR4. (This register is cleared to all 0s when entering SMM).
The hidden segment descriptor information stored in segment registers CS, DS,
ES, FS, GS, and SS.
Vol. 3B 26-7
SYSTEM MANAGEMENT MODE
If an SMI request is issued for the purpose of powering down the processor, the
values of all reserved locations in the SMM state save must be saved to nonvolatile
memory.
The following state is not automatically saved and restored following an SMI and the
RSM instruction, respectively:
•
•
•
•
•
Debug registers DR0 through DR3.
•
•
•
•
The state of the trap controller.
The x87 FPU registers.
The MTRRs.
Control register CR2.
The model-specific registers (for the P6 family and Pentium processors) or test
registers TR3 through TR7 (for the Pentium and Intel486 processors).
The machine-check architecture registers.
The APIC internal interrupt state (ISR, IRR, etc.).
The microcode update state.
If an SMI is used to power down the processor, a power-on reset will be required
before returning to SMM, which will reset much of this state back to its default
values. So an SMI handler that is going to trigger power down should first read these
registers listed above directly, and save them (along with the rest of RAM) to nonvolatile storage. After the power-on reset, the continuation of the SMI handler should
restore these values, along with the rest of the system's state. Anytime the SMI
handler changes these registers in the processor, it must also save and restore them.
NOTES
A small subset of the MSRs (such as, the time-stamp counter and
performance-monitoring counters) are not arbitrarily writable and
therefore cannot be saved and restored. SMM-based power-down
and restoration should only be performed with operating systems
that do not use or rely on the values of these registers.
Operating system developers should be aware of this fact and insure
that their operating-system assisted power-down and restoration
software is immune to unexpected changes in these register values.
26.4.1.1
SMRAM State Save Map and Intel 64 Architecture
When the processor initially enters SMM, it writes its state to the state save area of
the SMRAM. The state save area on an Intel 64 processor at [SMBASE + 8000H +
7FFFH] and extends to [SMBASE + 8000H + 7C00H].
Support for Intel 64 architecture is reported by CPUID.80000001:EDX[29] = 1. The
layout of the SMRAM state save map is shown in Table 26-3.
26-8 Vol. 3B
SYSTEM MANAGEMENT MODE
Additionally, the SMRAM state save map shown in Table 26-3 also applies to processors with the following CPUID signatures listed in Table 26-2, irrespective of the value
in CPUID.80000001:EDX[29].
Table 26-2. Processor Signatures and 64-bit SMRAM State Save Map Format
DisplayFamily_DisplayModel Processor Families/Processor Number Series
06_17H
Intel Xeon Processor 5200, 5400 series, Intel Core 2 Quad
processor Q9xxx, Intel Core 2 Duo processors E8000, T9000,
06_0FH
Intel Xeon Processor 3000, 3200, 5100, 5300, 7300 series, Intel
Core 2 Quad, Intel Core 2 Extreme, Intel Core 2 Duo processors,
Intel Pentium dual-core processors
06_1CH
Intel® Atom™ processors
Table 26-3. SMRAM State Save Map for Intel 64 Architecture
Offset
(Added to SMBASE +
8000H)
Register
Writable?
7FF8H
CR0
No
7FF0H
CR3
No
7FE8H
RFLAGS
Yes
7FE0H
IA32_EFER
Yes
7FD8H
RIP
Yes
7FD0H
DR6
No
7FC8H
DR7
No
7FC4H
TR SEL1
No
1
7FC0H
LDTR SEL
No
7FBCH
1
GS SEL
No
7FB8H
FS SEL1
No
7FB4H
1
No
7FB0H
SS SEL
1
No
7FACH
CS SEL1
No
7FA8H
ES SEL
1
No
7FA4H
IO_MISC
No
7F9CH
IO_MEM_ADDR
No
DS SEL
Vol. 3B 26-9
SYSTEM MANAGEMENT MODE
Table 26-3. SMRAM State Save Map for Intel 64 Architecture (Contd.)
Offset
(Added to SMBASE +
8000H)
Register
Writable?
7F94H
RDI
Yes
7F8CH
RSI
Yes
7F84H
RBP
Yes
7F7CH
RSP
Yes
7F74H
RBX
Yes
7F6CH
RDX
Yes
7F64H
RCX
Yes
7F5CH
RAX
Yes
7F54H
R8
Yes
7F4CH
R9
Yes
7F44H
R10
Yes
7F3CH
R11
Yes
7F34H
R12
Yes
7F2CH
R13
Yes
7F24H
R14
Yes
7F1CH
R15
Yes
7F1BH-7F04H
Reserved
No
7F02H
Auto HALT Restart Field (Word)
Yes
7F00H
I/O Instruction Restart Field (Word)
Yes
7EFCH
SMM Revision Identifier Field (Doubleword)
No
7EF8H
SMBASE Field (Doubleword)
Yes
7EF7H - 7EE4H
Reserved
No
7EE0H
Setting of “enable EPT” VM-execution control
No
7ED8H
Value of EPTP VM-execution control field
No
7ED7H - 7EA0H
Reserved
No
7E9CH
LDT Base (lower 32 bits)
No
7E98H
Reserved
No
7E94H
IDT Base (lower 32 bits)
No
7E90H
Reserved
No
26-10 Vol. 3B
SYSTEM MANAGEMENT MODE
Table 26-3. SMRAM State Save Map for Intel 64 Architecture (Contd.)
Offset
(Added to SMBASE +
8000H)
Register
Writable?
7E8CH
GDT Base (lower 32 bits)
No
7E8BH - 7E44H
Reserved
No
7E40H
CR4
No
7E3FH - 7DF0H
Reserved
No
7DE8H
IO_EIP
Yes
7DE7H - 7DDCH
Reserved
No
7DD8H
IDT Base (Upper 32 bits)
No
7DD4H
LDT Base (Upper 32 bits)
No
7DD0H
GDT Base (Upper 32 bits)
No
7DCFH - 7C00H
Reserved
No
NOTE:
1. The two most significant bytes are reserved.
26.4.2
SMRAM Caching
An IA-32 processor does not automatically write back and invalidate its caches before
entering SMM or before exiting SMM. Because of this behavior, care must be taken in
the placement of the SMRAM in system memory and in the caching of the SMRAM to
prevent cache incoherence when switching back and forth between SMM and
protected mode operation. Either of the following three methods of locating the
SMRAM in system memory will guarantee cache coherency:
•
Place the SRAM in a dedicated section of system memory that the operating
system and applications are prevented from accessing. Here, the SRAM can be
designated as cacheable (WB, WT, or WC) for optimum processor performance,
without risking cache incoherence when entering or exiting SMM.
•
Place the SRAM in a section of memory that overlaps an area used by the
operating system (such as the video memory), but designate the SMRAM as
uncacheable (UC). This method prevents cache access when in SMM to maintain
cache coherency, but the use of uncacheable memory reduces the performance
of SMM code.
•
Place the SRAM in a section of system memory that overlaps an area used by the
operating system and/or application code, but explicitly flush (write back and
invalidate) the caches upon entering and exiting SMM mode. This method
maintains cache coherency, but the incurs the overhead of two complete cache
flushes.
Vol. 3B 26-11
SYSTEM MANAGEMENT MODE
For Pentium 4, Intel Xeon, and P6 family processors, a combination of the first two
methods of locating the SMRAM is recommended. Here the SMRAM is split between
an overlapping and a dedicated region of memory. Upon entering SMM, the SMRAM
space that is accessed overlaps video memory (typically located in low memory).
This SMRAM section is designated as UC memory. The initial SMM code then jumps to
a second SMRAM section that is located in a dedicated region of system memory
(typically in high memory). This SMRAM section can be cached for optimum
processor performance.
For systems that explicitly flush the caches upon entering SMM (the third method
described above), the cache flush can be accomplished by asserting the FLUSH# pin
at the same time as the request to enter SMM (generally initiated by asserting the
SMI# pin). The priorities of the FLUSH# and SMI# pins are such that the FLUSH# is
serviced first. To guarantee this behavior, the processor requires that the following
constraints on the interaction of FLUSH# and SMI# be met. In a system where the
FLUSH# and SMI# pins are synchronous and the set up and hold times are met, then
the FLUSH# and SMI# pins may be asserted in the same clock. In asynchronous
systems, the FLUSH# pin must be asserted at least one clock before the SMI# pin to
guarantee that the FLUSH# pin is serviced first.
Upon leaving SMM (for systems that explicitly flush the caches), the WBINVD instruction should be executed prior to leaving SMM to flush the caches.
NOTES
In systems based on the Pentium processor that use the FLUSH# pin
to write back and invalidate cache contents before entering SMM, the
processor will prefetch at least one cache line in between when the
Flush Acknowledge cycle is run and the subsequent recognition of
SMI# and the assertion of SMIACT#.
It is the obligation of the system to ensure that these lines are not
cached by returning KEN# inactive to the Pentium processor.
26.4.2.1
System Management Range Registers (SMRR)
SMI handler code and data stored by SMM code resides in SMRAM. The SMRR interface is an enhancement in Intel 64 architecture to limit cacheable reference of
addresses in SMRAM to code running in SMM. The SMRR interface can be configured
only by code running in SMM. Details of SMRR is described in Section 11.11.2.4.
26.5
SMI HANDLER EXECUTION ENVIRONMENT
After saving the current context of the processor, the processor initializes its core
registers to the values shown in Table 26-4. Upon entering SMM, the PE and PG flags
in control register CR0 are cleared, which places the processor is in an environment
similar to real-address mode. The differences between the SMM execution environment and the real-address mode execution environment are as follows:
26-12 Vol. 3B
SYSTEM MANAGEMENT MODE
•
The addressable SMRAM address space ranges from 0 to FFFFFFFFH (4 GBytes).
(The physical address extension — enabled with the PAE flag in control register
CR4 — is not supported in SMM.)
•
The normal 64-KByte segment limit for real-address mode is increased to
4 GBytes.
•
The default operand and address sizes are set to 16 bits, which restricts the
addressable SMRAM address space to the 1-MByte real-address mode limit for
native real-address-mode code. However, operand-size and address-size
override prefixes can be used to access the address space beyond the 1-MByte.
Table 26-4. Processor Register Initialization in SMM
Register
Contents
General-purpose registers
Undefined
EFLAGS
00000002H
EIP
00008000H
CS selector
SMM Base shifted right 4 bits (default 3000H)
CS base
SMM Base (default 30000H)
DS, ES, FS, GS, SS Selectors
0000H
DS, ES, FS, GS, SS Bases
000000000H
DS, ES, FS, GS, SS Limits
0FFFFFFFFH
CR0
PE, EM, TS, and PG flags set to 0; others unmodified
CR4
Cleared to zero
DR6
Undefined
DR7
00000400H
•
Near jumps and calls can be made to anywhere in the 4-GByte address space if a
32-bit operand-size override prefix is used. Due to the real-address-mode style
of base-address formation, a far call or jump cannot transfer control to a
segment with a base address of more than 20 bits (1 MByte). However, since the
segment limit in SMM is 4 GBytes, offsets into a segment that go beyond the
1-MByte limit are allowed when using 32-bit operand-size override prefixes. Any
program control transfer that does not have a 32-bit operand-size override prefix
truncates the EIP value to the 16 low-order bits.
•
Data and the stack can be located anywhere in the 4-GByte address space, but
can be accessed only with a 32-bit address-size override if they are located above
1 MByte. As with the code segment, the base address for a data or stack segment
cannot be more than 20 bits.
The value in segment register CS is automatically set to the default of 30000H for the
SMBASE shifted 4 bits to the right; that is, 3000H. The EIP register is set to 8000H.
When the EIP value is added to shifted CS value (the SMBASE), the resulting linear
address points to the first instruction of the SMI handler.
Vol. 3B 26-13
SYSTEM MANAGEMENT MODE
The other segment registers (DS, SS, ES, FS, and GS) are cleared to 0 and their
segment limits are set to 4 GBytes. In this state, the SMRAM address space may be
treated as a single flat 4-GByte linear address space. If a segment register is loaded
with a 16-bit value, that value is then shifted left by 4 bits and loaded into the
segment base (hidden part of the segment register). The limits and attributes are not
modified.
Maskable hardware interrupts, exceptions, NMI interrupts, SMI interrupts, A20M
interrupts, single-step traps, breakpoint traps, and INIT operations are inhibited
when the processor enters SMM. Maskable hardware interrupts, exceptions, singlestep traps, and breakpoint traps can be enabled in SMM if the SMM execution environment provides and initializes an interrupt table and the necessary interrupt and
exception handlers (see Section 26.6).
26.6
EXCEPTIONS AND INTERRUPTS WITHIN SMM
When the processor enters SMM, all hardware interrupts are disabled in the following
manner:
•
The IF flag in the EFLAGS register is cleared, which inhibits maskable hardware
interrupts from being generated.
•
•
The TF flag in the EFLAGS register is cleared, which disables single-step traps.
•
NMI, SMI, and A20M interrupts are blocked by internal SMM logic. (See Section
26.8 for more information about how NMIs are handled in SMM.)
Debug register DR7 is cleared, which disables breakpoint traps. (This action
prevents a debugger from accidentally breaking into an SMM handler if a debug
breakpoint is set in normal address space that overlays code or data in SMRAM.)
Software-invoked interrupts and exceptions can still occur, and maskable hardware
interrupts can be enabled by setting the IF flag. Intel recommends that SMM code be
written in so that it does not invoke software interrupts (with the INT n, INTO, INT 3,
or BOUND instructions) or generate exceptions.
If the SMM handler requires interrupt and exception handling, an SMM interrupt table
and the necessary exception and interrupt handlers must be created and initialized
from within SMM. Until the interrupt table is correctly initialized (using the LIDT
instruction), exceptions and software interrupts will result in unpredictable processor
behavior.
The following restrictions apply when designing SMM interrupt and exceptionhandling facilities:
•
The interrupt table should be located at linear address 0 and must contain realaddress mode style interrupt vectors (4 bytes containing CS and IP).
•
Due to the real-address mode style of base address formation, an interrupt or
exception cannot transfer control to a segment with a base address of more that
20 bits.
26-14 Vol. 3B
SYSTEM MANAGEMENT MODE
•
An interrupt or exception cannot transfer control to a segment offset of more
than 16 bits (64 KBytes).
•
When an exception or interrupt occurs, only the 16 least-significant bits of the
return address (EIP) are pushed onto the stack. If the offset of the interrupted
procedure is greater than 64 KBytes, it is not possible for the interrupt/exception
handler to return control to that procedure. (One solution to this problem is for a
handler to adjust the return address on the stack.)
•
The SMBASE relocation feature affects the way the processor will return from an
interrupt or exception generated while the SMI handler is executing. For
example, if the SMBASE is relocated to above 1 MByte, but the exception
handlers are below 1 MByte, a normal return to the SMI handler is not possible.
One solution is to provide the exception handler with a mechanism for calculating
a return address above 1 MByte from the 16-bit return address on the stack, then
use a 32-bit far call to return to the interrupted procedure.
•
If an SMI handler needs access to the debug trap facilities, it must insure that an
SMM accessible debug handler is available and save the current contents of
debug registers DR0 through DR3 (for later restoration). Debug registers DR0
through DR3 and DR7 must then be initialized with the appropriate values.
•
If an SMI handler needs access to the single-step mechanism, it must insure that
an SMM accessible single-step handler is available, and then set the TF flag in the
EFLAGS register.
•
If the SMI design requires the processor to respond to maskable hardware
interrupts or software-generated interrupts while in SMM, it must ensure that
SMM accessible interrupt handlers are available and then set the IF flag in the
EFLAGS register (using the STI instruction). Software interrupts are not blocked
upon entry to SMM, so they do not need to be enabled.
26.7
MANAGING SYNCHRONOUS AND ASYNCHRONOUS
SYSTEM MANAGEMENT INTERRUPTS
When coding for a multiprocessor system or a system with Intel HT Technology, it
was not always possible for an SMI handler to distinguish between a synchronous
SMI (triggered during an I/O instruction) and an asynchronous SMI. To facilitate the
discrimination of these two events, incremental state information has been added to
the SMM state save map.
Processors that have an SMM revision ID of 30004H or higher have the incremental
state information described below.
26.7.1
I/O State Implementation
Within the extended SMM state save map, a bit (IO_SMI) is provided that is set only
when an SMI is either taken immediately after a successful I/O instruction or is taken
Vol. 3B 26-15
SYSTEM MANAGEMENT MODE
after a successful iteration of a REP I/O instruction (the successful notion pertains to
the processor point of view; not necessarily to the corresponding platform function).
When set, the IO_SMI bit provides a strong indication that the corresponding SMI
was synchronous. In this case, the SMM State Save Map also supplies the port
address of the I/O operation. The IO_SMI bit and the I/O Port Address may be used
in conjunction with the information logged by the platform to confirm that the SMI
was indeed synchronous.
The IO_SMI bit by itself is a strong indication, not a guarantee, that the SMI is
synchronous. This is because an asynchronous SMI might coincidentally be taken
after an I/O instruction. In such a case, the IO_SMI bit would still be set in the SMM
state save map.
Information characterizing the I/O instruction is saved in two locations in the SMM
State Save Map (Table 26-5). The IO_SMI bit also serves as a valid bit for the rest of
the I/O information fields. The contents of these I/O information fields are not
defined when the IO_SMI bit is not set.
Table 26-5. I/O Instruction Information in the SMM State Save Map
State (SMM Rev. ID: 30004H or
higher)
Format
31
15
7
4
3
I/O Memory Address Field
0
I/O Memory Address
SMRAM offset 7FA0
When IO_SMI is set, the other fields may be interpreted as follows:
•
•
I/O length:
•
•
•
001 – Byte
010 – Word
100 – Dword
I/O instruction type (Table 26-6)
Table 26-6. I/O Instruction Type Encodings
Instruction
Encoding
IN Immediate
1001
IN DX
0001
OUT Immediate
1000
26-16 Vol. 3B
0
IO_SMI
31
1
I/O Length
I/O Type
SMRAM offset 7FA4
8
Reserved
I/O Port
I/0 State Field
16
SYSTEM MANAGEMENT MODE
Table 26-6. I/O Instruction Type Encodings (Contd.)
Instruction
Encoding
OUT DX
0000
INS
0011
OUTS
0010
REP INS
0111
REP OUTS
0110
26.8
NMI HANDLING WHILE IN SMM
NMI interrupts are blocked upon entry to the SMI handler. If an NMI request occurs
during the SMI handler, it is latched and serviced after the processor exits SMM. Only
one NMI request will be latched during the SMI handler. If an NMI request is pending
when the processor executes the RSM instruction, the NMI is serviced before the next
instruction of the interrupted code sequence. This assumes that NMIs were not
blocked before the SMI occurred. If NMIs were blocked before the SMI occurred, they
are blocked after execution of RSM.
Although NMI requests are blocked when the processor enters SMM, they may be
enabled through software by executing an IRET instruction. If the SMM handler
requires the use of NMI interrupts, it should invoke a dummy interrupt service
routine for the purpose of executing an IRET instruction. Once an IRET instruction is
executed, NMI interrupt requests are serviced in the same “real mode” manner in
which they are handled outside of SMM.
A special case can occur if an SMI handler nests inside an NMI handler and then
another NMI occurs. During NMI interrupt handling, NMI interrupts are disabled, so
normally NMI interrupts are serviced and completed with an IRET instruction one at
a time. When the processor enters SMM while executing an NMI handler, the
processor saves the SMRAM state save map but does not save the attribute to keep
NMI interrupts disabled. Potentially, an NMI could be latched (while in SMM or upon
exit) and serviced upon exit of SMM even though the previous NMI handler has still
not completed. One or more NMIs could thus be nested inside the first NMI handler.
The NMI interrupt handler should take this possibility into consideration.
Also, for the Pentium processor, exceptions that invoke a trap or fault handler will
enable NMI interrupts from inside of SMM. This behavior is implementation specific
for the Pentium processor and is not part of the IA-32 architecture.
26.9
SMM REVISION IDENTIFIER
The SMM revision identifier field is used to indicate the version of SMM and the SMM
extensions that are supported by the processor (see Figure 26-2). The SMM revision
identifier is written during SMM entry and can be examined in SMRAM space at offset
Vol. 3B 26-17
SYSTEM MANAGEMENT MODE
7EFCH. The lower word of the SMM revision identifier refers to the version of the base
SMM architecture.
Register Offset
7EFCH
31
0
18 17 16 15
Reserved
SMM Revision Identifier
SMBASE Relocation
I/O Instruction Restart
Figure 26-2. SMM Revision Identifier
The upper word of the SMM revision identifier refers to the extensions available. If
the I/O instruction restart flag (bit 16) is set, the processor supports the I/O instruction restart (see Section 26.12); if the SMBASE relocation flag (bit 17) is set, SMRAM
base address relocation is supported (see Section 26.11).
26.10
AUTO HALT RESTART
If the processor is in a HALT state (due to the prior execution of a HLT instruction)
when it receives an SMI, the processor records the fact in the auto HALT restart flag
in the saved processor state (see Figure 26-3). (This flag is located at offset 7F02H
and bit 0 in the state save area of the SMRAM.)
If the processor sets the auto HALT restart flag upon entering SMM (indicating that
the SMI occurred when the processor was in the HALT state), the SMI handler has
two options:
•
It can leave the auto HALT restart flag set, which instructs the RSM instruction to
return program control to the HLT instruction. This option in effect causes the
processor to re-enter the HALT state after handling the SMI. (This is the default
operation.)
•
It can clear the auto HALT restart flag, with instructs the RSM instruction to
return program control to the instruction following the HLT instruction.
26-18 Vol. 3B
SYSTEM MANAGEMENT MODE
1 0
15
Reserved
Register Offset
7F02H
Auto HALT Restart
Figure 26-3. Auto HALT Restart Field
These options are summarized in Table 26-7. If the processor was not in a HALT state
when the SMI was received (the auto HALT restart flag is cleared), setting the flag to
1 will cause unpredictable behavior when the RSM instruction is executed.
Table 26-7. Auto HALT Restart Flag Values
Value of Flag
After Entry to
SMM
Value of Flag
Action of Processor When Exiting SMM
When Exiting SMM
0
0
Returns to next instruction in interrupted program or task.
0
1
Unpredictable.
1
0
Returns to next instruction after HLT instruction.
1
1
Returns to HALT state.
If the HLT instruction is restarted, the processor will generate a memory access to
fetch the HLT instruction (if it is not in the internal cache), and execute a HLT bus
transaction. This behavior results in multiple HLT bus transactions for the same HLT
instruction.
26.10.1 Executing the HLT Instruction in SMM
The HLT instruction should not be executed during SMM, unless interrupts have been
enabled by setting the IF flag in the EFLAGS register. If the processor is halted in
SMM, the only event that can remove the processor from this state is a maskable
hardware interrupt or a hardware reset.
26.11
SMBASE RELOCATION
The default base address for the SMRAM is 30000H. This value is contained in an
internal processor register called the SMBASE register. The operating system or
executive can relocate the SMRAM by setting the SMBASE field in the saved state
map (at offset 7EF8H) to a new value (see Figure 26-4). The RSM instruction reloads
the internal SMBASE register with the value in the SMBASE field each time it exits
SMM. All subsequent SMI requests will use the new SMBASE value to find the starting
Vol. 3B 26-19
SYSTEM MANAGEMENT MODE
address for the SMI handler (at SMBASE + 8000H) and the SMRAM state save area
(from SMBASE + FE00H to SMBASE + FFFFH). (The processor resets the value in its
internal SMBASE register to 30000H on a RESET, but does not change it on an INIT.)
31
0
SMM Base
Register Offset
7EF8H
Figure 26-4. SMBASE Relocation Field
In multiple-processor systems, initialization software must adjust the SMBASE value
for each processor so that the SMRAM state save areas for each processor do not
overlap. (For Pentium and Intel486 processors, the SMBASE values must be aligned
on a 32-KByte boundary or the processor will enter shutdown state during the execution of a RSM instruction.)
If the SMBASE relocation flag in the SMM revision identifier field is set, it indicates the
ability to relocate the SMBASE (see Section 26.9).
26.11.1 Relocating SMRAM to an Address Above 1 MByte
In SMM, the segment base registers can only be updated by changing the value in the
segment registers. The segment registers contain only 16 bits, which allows only 20
bits to be used for a segment base address (the segment register is shifted left 4 bits
to determine the segment base address). If SMRAM is relocated to an address above
1 MByte, software operating in real-address mode can no longer initialize the
segment registers to point to the SMRAM base address (SMBASE).
The SMRAM can still be accessed by using 32-bit address-size override prefixes to
generate an offset to the correct address. For example, if the SMBASE has been relocated to FFFFFFH (immediately below the 16-MByte boundary) and the DS, ES, FS,
and GS registers are still initialized to 0H, data in SMRAM can be accessed by using
32-bit displacement registers, as in the following example:
mov
mov
esi,00FFxxxxH; 64K segment immediately below 16M
ax,ds:[esi]
A stack located above the 1-MByte boundary can be accessed in the same manner.
26.12
I/O INSTRUCTION RESTART
If the I/O instruction restart flag in the SMM revision identifier field is set (see Section
26.9), the I/O instruction restart mechanism is present on the processor. This mechanism allows an interrupted I/O instruction to be re-executed upon returning from
26-20 Vol. 3B
SYSTEM MANAGEMENT MODE
SMM mode. For example, if an I/O instruction is used to access a powered-down I/O
device, a chip set supporting this device can intercept the access and respond by
asserting SMI#. This action invokes the SMI handler to power-up the device. Upon
returning from the SMI handler, the I/O instruction restart mechanism can be used to
re-execute the I/O instruction that caused the SMI.
The I/O instruction restart field (at offset 7F00H in the SMM state-save area, see
Figure 26-5) controls I/O instruction restart. When an RSM instruction is executed, if
this field contains the value FFH, then the EIP register is modified to point to the I/O
instruction that received the SMI request. The processor will then automatically reexecute the I/O instruction that the SMI trapped. (The processor saves the necessary
machine state to insure that re-execution of the instruction is handled coherently.)
15
0
I/O Instruction Restart Field
Register Offset
7F00H
Figure 26-5. I/O Instruction Restart Field
If the I/O instruction restart field contains the value 00H when the RSM instruction is
executed, then the processor begins program execution with the instruction following
the I/O instruction. (When a repeat prefix is being used, the next instruction may be
the next I/O instruction in the repeat loop.) Not re-executing the interrupted I/O
instruction is the default behavior; the processor automatically initializes the I/O
instruction restart field to 00H upon entering SMM. Table 26-8 summarizes the states
of the I/O instruction restart field.
Table 26-8. I/O Instruction Restart Field Values
Value of Flag After
Entry to SMM
Value of Flag When
Exiting SMM
Action of Processor When Exiting SMM
00H
00H
Does not re-execute trapped I/O instruction.
00H
FFH
Re-executes trapped I/O instruction.
The I/O instruction restart mechanism does not indicate the cause of the SMI. It is
the responsibility of the SMI handler to examine the state of the processor to determine the cause of the SMI and to determine if an I/O instruction was interrupted and
should be restarted upon exiting SMM. If an SMI interrupt is signaled on a non-I/O
instruction boundary, setting the I/O instruction restart field to FFH prior to executing
the RSM instruction will likely result in a program error.
Vol. 3B 26-21
SYSTEM MANAGEMENT MODE
26.12.1 Back-to-Back SMI Interrupts When I/O Instruction Restart Is
Being Used
If an SMI interrupt is signaled while the processor is servicing an SMI interrupt that
occurred on an I/O instruction boundary, the processor will service the new SMI
request before restarting the originally interrupted I/O instruction. If the I/O instruction restart field is set to FFH prior to returning from the second SMI handler, the EIP
will point to an address different from the originally interrupted I/O instruction, which
will likely lead to a program error. To avoid this situation, the SMI handler must be
able to recognize the occurrence of back-to-back SMI interrupts when I/O instruction
restart is being used and insure that the handler sets the I/O instruction restart field
to 00H prior to returning from the second invocation of the SMI handler.
26.13
SMM MULTIPLE-PROCESSOR CONSIDERATIONS
The following should be noted when designing multiple-processor systems:
•
•
Any processor in a multiprocessor system can respond to an SMM.
•
The SMRAMs for different processors can be overlapped in the same memory
space. The only stipulation is that each processor needs its own state save area
and its own dynamic data storage area. (Also, for the Pentium and Intel486
processors, the SMBASE address must be located on a 32-KByte boundary.) Code
and static data can be shared among processors. Overlapping SMRAM spaces can
be done more efficiently with the P6 family processors because they do not
require that the SMBASE address be on a 32-KByte boundary.
•
•
The SMI handler will need to initialize the SMBASE for each processor.
•
•
Two or more processors can be executing in SMM at the same time.
Each processor needs its own SMRAM space. This space can be in system
memory or in a separate RAM.
Processors can respond to local SMIs through their SMI# pins or to SMIs received
through the APIC interface. The APIC interface can distribute SMIs to different
processors.
When operating Pentium processors in dual processing (DP) mode, the SMIACT#
pin is driven only by the MRM processor and should be sampled with ADS#. For
additional details, see Chapter 14 of the Pentium Processor Family User’s Manual,
Volume 1.
SMM is not re-entrant, because the SMRAM State Save Map is fixed relative to the
SMBASE. If there is a need to support two or more processors in SMM mode at the
same time then each processor should have dedicated SMRAM spaces. This can be
done by using the SMBASE Relocation feature (see Section 26.11).
26-22 Vol. 3B
SYSTEM MANAGEMENT MODE
26.14
DEFAULT TREATMENT OF SMIS AND SMM WITH VMX
OPERATION AND SMX OPERATION
Under the default treatment, the interactions of SMIs and SMM with VMX operation
are few. This section details those interactions. It also explains how this treatment
affects SMX operation.
26.14.1 Default Treatment of SMI Delivery
Ordinary SMI delivery saves processor state into SMRAM and then loads state based
on architectural definitions. Under the default treatment, processors that support
VMX operation perform SMI delivery as follows:
enter SMM;
save the following internal to the processor:
CR4.VMXE
an indication of whether the logical processor was in VMX operation (root or non-root)
IF the logical processor is in VMX operation
THEN
save current VMCS pointer internal to the processor;
leave VMX operation;
save VMX-critical state defined below;
FI;
IF the logical processor supports SMX operation
THEN
save internal to the logical processor an indication of whether the Intel® TXT private space
is locked;
IF the TXT private space is unlocked
THEN lock the TXT private space;
FI;
FI;
CR4.VMXE ← 0;
perform ordinary SMI delivery:
save processor state in SMRAM;
set processor state to standard SMM values;1
invalidate linear mappings and combined mappings associated with VPID 0000H (for all PCIDs);
combined mappings for VPID 0000H are invalidated for all EP4TA values (EP4TA is the value of bits
51:12 of EPTP; see Section 25.3);
The pseudocode above makes reference to the saving of VMX-critical state. This
state consists of the following: (1) SS.DPL (the current privilege level);
(2) RFLAGS.VM2; (3) the state of blocking by STI and by MOV SS (see Table 21-3 in
1. This causes the logical processor to block INIT signals, NMIs, and SMIs.
Vol. 3B 26-23
SYSTEM MANAGEMENT MODE
Section 21.4.2); (4) the state of virtual-NMI blocking (only if the processor is in VMX
non-root operation and the “virtual NMIs” VM-execution control is 1); and (5) an
indication of whether an MTF VM exit is pending (see Section 22.7.2). These data
may be saved internal to the processor or in the VMCS region of the current VMCS.
Processors that do not support SMI recognition while there is blocking by STI or by
MOV SS need not save the state of such blocking.
If the logical processor supports the 1-setting of the “enable EPT” VM-execution
control and the logical processor was in VMX non-root operation at the time of an
SMI, it saves the value of that control into bit 0 of the 32-bit field at offset SMBASE +
8000H + 7EE0H (SMBASE + FEE0H; see Table 26-3).1 If the logical processor was
not in VMX non-root operation at the time of the SMI, it saves 0 into that bit. If the
logical processor saves 1 into that bit (it was in VMX non-root operation and the
“enable EPT” VM-execution control was 1), it saves the value of the EPT pointer
(EPTP) into the 64-bit field at offset SMBASE + 8000H + 7ED8H (SMBASE + FED8H).
Because SMI delivery causes a logical processor to leave VMX operation, all the
controls associated with VMX non-root operation are disabled in SMM and thus
cannot cause VM exits while the logical processor in SMM.
26.14.2 Default Treatment of RSM
Ordinary execution of RSM restores processor state from SMRAM. Under the default
treatment, processors that support VMX operation perform RSM as follows:
IF VMXE = 1 in CR4 image in SMRAM
THEN fail and enter shutdown state;
ELSE
restore state normally from SMRAM;
invalidate linear mappings and combined mappings associated with all VPIDs and all PCIDs;
combined mappings are invalidated for all EP4TA values (EP4TA is the value of bits 51:12 of EPTP;
see Section 25.3);
IF the logical processor supports SMX operation andthe Intel® TXT private space was
unlocked at the time of the last SMI (as saved)
THEN unlock the TXT private space;
FI;
CR4.VMXE ← value stored internally;
2. Section 26.14 and Section 26.15 use the notation RAX, RIP, RSP, RFLAGS, etc. for processor registers because most processors that support VMX operation also support Intel 64 architecture.
For processors that do not support Intel 64 architecture, this notation refers to the 32-bit forms
of these registers (EAX, EIP, ESP, EFLAGS, etc.). In a few places, notation such as EAX is used to
refer specifically to the lower 32 bits of the register.
1. “Enable EPT” is a secondary processor-based VM-execution control. If bit 31 of the primary processor-based VM-execution controls is 0, SMI functions as the “enable EPT” VM-execution control
were 0. See Section 21.6.2.
26-24 Vol. 3B
SYSTEM MANAGEMENT MODE
IF internal storage indicates that the logical processor
had been in VMX operation (root or non-root)
THEN
enter VMX operation (root or non-root);
restore VMX-critical state as defined in Section 26.14.1;
set to their fixed values any bits in CR0 and CR4 whose values must be fixed in
VMX operation (see Section 20.8);1
IF RFLAGS.VM = 0 AND (in VMX root operation OR the “unrestricted guest” VMexecution control is 0)2
THEN
CS.RPL ← SS.DPL;
SS.RPL ← SS.DPL;
FI;
restore current VMCS pointer;
FI;
leave SMM;
IF logical processor will be in VMX operation or in SMX operation after RSM
THEN block A20M and leave A20M mode;
FI;
FI;
RSM unblocks SMIs. It restores the state of blocking by NMI (see Table 21-3 in
Section 21.4.2) as follows:
•
If the RSM is not to VMX non-root operation or if the “virtual NMIs” VM-execution
control will be 0, the state of NMI blocking is restored normally.
•
If the RSM is to VMX non-root operation and the “virtual NMIs” VM-execution
control will be 1, NMIs are not blocked after RSM. The state of virtual-NMI
blocking is restored as part of VMX-critical state.
INIT signals are blocked after RSM if and only if the logical processor will be in VMX
root operation.
If RSM returns a logical processor to VMX non-root operation, it re-establishes the
controls associated with the current VMCS. If the “interrupt-window exiting”
VM-execution control is 1, a VM exit occurs immediately after RSM if the enabling
conditions apply. The same is true for the “NMI-window exiting” VM-execution
control. Such VM exits occur with their normal priority. See Section 22.3.
1. If the RSM is to VMX non-root operation and both the “unrestricted guest” VM-execution control
and bit 31 of the primary processor-based VM-execution controls will be 1, CR0.PE and CR0.PG
retain the values that were loaded from SMRAM regardless of what is reported in the capability
MSR IA32_VMX_CR0_FIXED0.
2. “Unrestricted guest” is a secondary processor-based VM-execution control. If bit 31 of the primary processor-based VM-execution controls is 0, VM entry functions as if the “unrestricted
guest” VM-execution control were 0. See Section 21.6.2.
Vol. 3B 26-25
SYSTEM MANAGEMENT MODE
If an MTF VM exit was pending at the time of the previous SMI, an MTF VM exit is
pending on the instruction boundary following execution of RSM. The following items
detail the treatment of MTF VM exits that may be pending following RSM:
•
System-management interrupts (SMIs), INIT signals, and higher priority events
take priority over these MTF VM exits. These MTF VM exits take priority over
debug-trap exceptions and lower priority events.
•
These MTF VM exits wake the logical processor if RSM caused the logical
processor to enter the HLT state (see Section 26.10). They do not occur if the
logical processor just entered the shutdown state.
26.14.3 Protection of CR4.VMXE in SMM
Under the default treatment, CR4.VMXE is treated as a reserved bit while a logical
processor is in SMM. Any attempt by software running in SMM to set this bit causes a
general-protection exception. In addition, software cannot use VMX instructions or
enter VMX operation while in SMM.
26.14.4 VMXOFF and SMI Unblocking
The VMXOFF instruction can be executed only with the default treatment (see Section
26.15.1) and only outside SMM. If SMIs are blocked when VMXOFF is executed,
VMXOFF unblocks them unless IA32_SMM_MONITOR_CTL[bit 2] is 1 (see Section
26.15.5 for details regarding this MSR).1 Section 26.15.7 identifies a case in which
SMIs may be blocked when VMXOFF is executed.
Not all processors allow this bit to be set to 1. Software should consult the VMX capability MSR IA32_VMX_MISC (see Appendix G.6) to determine whether this is allowed.
26.15
DUAL-MONITOR TREATMENT OF SMIs AND SMM
Dual-monitor treatment is activated through the cooperation of the executive
monitor (the VMM that operates outside of SMM to provide basic virtualization) and
the SMM monitor (the VMM that operates inside SMM—while in VMX operation—to
support system-management functions). Control is transferred to the SMM monitor
through VM exits; VM entries are used to return from SMM.
The dual-monitor treatment may not be supported by all processors. Software should
consult the VMX capability MSR IA32_VMX_BASIC (see Appendix G.1) to determine
whether it is supported.
1. Setting IA32_SMM_MONITOR_CTL[bit 2] to 1 prevents VMXOFF from unblocking SMIs regardless
of the value of the register’s valid bit (bit 0).
26-26 Vol. 3B
SYSTEM MANAGEMENT MODE
26.15.1 Dual-Monitor Treatment Overview
The dual-monitor treatment uses an executive monitor and an SMM monitor. Transitions from the executive monitor or its guests to the SMM monitor are called SMM
VM exits and are discussed in Section 26.15.2. SMM VM exits are caused by SMIs as
well as executions of VMCALL in VMX root operation. The latter allow the executive
monitor to call the SMM monitor for service.
The SMM monitor runs in VMX root operation and uses VMX instructions to establish
a VMCS and perform VM entries to its own guests. This is done all inside SMM (see
Section 26.15.3). The SMM monitor returns from SMM, not by using the RSM instruction, but by using a VM entry that returns from SMM. Such VM entries are described
in Section 26.15.4.
Initially, there is no SMM monitor and the default treatment (Section 26.14) is used.
The dual-monitor treatment is not used until it is enabled and activated. The steps to
do this are described in Section 26.15.5 and Section 26.15.6.
It is not possible to leave VMX operation under the dual-monitor treatment; VMXOFF
will fail if executed. The dual-monitor treatment must be deactivated first. The SMM
monitor deactivates dual-monitor treatment using a VM entry that returns from SMM
with the “deactivate dual-monitor treatment” VM-entry control set to 1 (see Section
26.15.7).
The executive monitor configures any VMCS that it uses for VM exits to the executive
monitor. SMM VM exits, which transfer control to the SMM monitor, use a different
VMCS. Under the dual-monitor treatment, each logical processor uses a separate
VMCS called the SMM-transfer VMCS. When the dual-monitor treatment is active,
the logical processor maintains another VMCS pointer called the SMM-transfer
VMCS pointer. The SMM-transfer VMCS pointer is established when the dualmonitor treatment is activated.
26.15.2 SMM VM Exits
An SMM VM exit is a VM exit that begins outside SMM and that ends in SMM.
Unlike other VM exits, SMM VM exits can begin in VMX root operation. SMM VM exits
result from the arrival of an SMI outside SMM or from execution of VMCALL in VMX
root operation outside SMM. Execution of VMCALL in VMX root operation causes an
SMM VM exit only if the valid bit is set in the IA32_SMM_MONITOR_CTL MSR (see
Section 26.15.5).
Execution of VMCALL in VMX root operation causes an SMM VM exit even under the
default treatment. This SMM VM exit activates the dual-monitor treatment (see
Section 26.15.6).
Differences between SMM VM exits and other VM exits are detailed in Sections
26.15.2.1 through 26.15.2.5. Differences between SMM VM exits that activate the
dual-monitor treatment and other SMM VM exits are described in Section 26.15.6.
Vol. 3B 26-27
SYSTEM MANAGEMENT MODE
26.15.2.1 Architectural State Before a VM Exit
System-management interrupts (SMIs) that cause SMM VM exits always do so
directly. They do not save state to SMRAM as they do under the default treatment.
26.15.2.2 Updating the Current-VMCS and Executive-VMCS Pointers
SMM VM exits begin by performing the following steps:
1. The executive-VMCS pointer field in the SMM-transfer VMCS is loaded as follows:
— If the SMM VM exit commenced in VMX non-root operation, it receives the
current-VMCS pointer.
— If the SMM VM exit commenced in VMX root operation, it receives the VMXON
pointer.
2. The current-VMCS pointer is loaded with the value of the SMM-transfer VMCS
pointer.
The last step ensures that the current VMCS is the SMM-transfer VMCS. VM-exit
information is recorded in that VMCS, and VM-entry control fields in that VMCS are
updated. State is saved into the guest-state area of that VMCS. The VM-exit controls
and host-state area of that VMCS determine how the VM exit operates.
26.15.2.3 Recording VM-Exit Information
SMM VM exits differ from other VM exit with regard to the way they record VM-exit
information. The differences follow.
•
Exit reason.
— Bits 15:0 of this field contain the basic exit reason. The field is loaded with
the reason for the SMM VM exit: I/O SMI (an SMI arrived immediately after
retirement of an I/O instruction), other SMI, or VMCALL. See Appendix I,
“VMX Basic Exit Reasons”.
— SMM VM exits are the only VM exits that may occur in VMX root operation.
Because the SMM monitor may need to know whether it was invoked from
VMX root or VMX non-root operation, this information is stored in bit 29 of the
exit-reason field (see Table 21-13 in Section 21.9.1). The bit is set by SMM
VM exits from VMX root operation.
— If the SMM VM exit occurred in VMX non-root operation and an MTF VM exit
was pending, bit 28 of the exit-reason field is set; otherwise, it is cleared.
— Bits 27:16 and bits 31:30 are cleared.
•
Exit qualification. For an SMM VM exit due an SMI that arrives immediately
after the retirement of an I/O instruction, the exit qualification contains
information about the I/O instruction that retired immediately before the SMI.It
has the format given in Table 26-9.
26-28 Vol. 3B
SYSTEM MANAGEMENT MODE
Table 26-9. Exit Qualification for SMIs That Arrive Immediately
After the Retirement of an I/O Instruction
Bit Position(s)
Contents
2:0
Size of access:
0 = 1-byte
1 = 2-byte
3 = 4-byte
Other values not used.
3
Direction of the attempted access (0 = OUT, 1 = IN)
4
String instruction (0 = not string; 1 = string)
5
REP prefixed (0 = not REP; 1 = REP)
6
Operand encoding (0 = DX, 1 = immediate)
15:7
Reserved (cleared to 0)
31:16
Port number (as specified in the I/O instruction)
63:32
Reserved (cleared to 0). These bits exist only on processors
that support Intel 64 architecture.
•
Guest linear address. This field is used for VM exits due to SMIs that arrive
immediately after the retirement of an INS or OUTS instruction for which the
relevant segment (ES for INS; DS for OUTS unless overridden by an instruction
prefix) is usable. The field receives the value of the linear address generated by
ES:(E)DI (for INS) or segment:(E)SI (for OUTS; the default segment is DS but
can be overridden by a segment override prefix) at the time the instruction
started. If the relevant segment is not usable, the value is undefined. On
processors that support Intel 64 architecture, bits 63:32 are clear if the logical
processor was not in 64-bit mode before the VM exit.
•
I/O RCX, I/O RSI, I/O RDI, and I/O RIP. For an SMM VM exit due an SMI
that arrives immediately after the retirement of an I/O instruction, these fields
receive the values that were in RCX, RSI, RDI, and RIP, respectively, before the
I/O instruction executed. Thus, the value saved for I/O RIP addresses the I/O
instruction.
26.15.2.4 Saving Guest State
SMM VM exits save the contents of the SMBASE register into the corresponding field
in the guest-state area.
Vol. 3B 26-29
SYSTEM MANAGEMENT MODE
The value of the VMX-preemption timer is saved into the corresponding field in the
guest-state area if the “save VMX-preemption timer value” VM-exit control is 1. That
field becomes undefined if, in addition, either the SMM VM exit is from VMX root
operation or the SMM VM exit is from VMX non-root operation and the “activate VMXpreemption timer” VM-execution control is 0.
26.15.2.5 Updating Non-Register State
SMM VM exits affect the non-register state of a logical processor as follows:
•
SMM VM exits cause non-maskable interrupts (NMIs) to be blocked; they may be
unblocked through execution of IRET or through a VM entry (depending on the
value loaded for the interruptibility state and the setting of the “virtual NMIs”
VM-execution control).
•
SMM VM exits cause SMIs to be blocked; they may be unblocked by a VM entry
that returns from SMM (see Section 26.15.4).
SMM VM exits invalidate linear mappings and combined mappings associated with
VPID 0000H for all PCIDs. Combined mappings for VPID 0000H are invalidated for all
EP4TA values (EP4TA is the value of bits 51:12 of EPTP; see Section 25.3). (Ordinary
VM exits are not required to perform such invalidation if the “enable VPID” VMexecution control is 1; see Section 24.5.5.)
26.15.3 Operation of an SMM Monitor
Once invoked, an SMM monitor is in VMX root operation and can use VMX instructions
to configure VMCSs and to cause VM entries to virtual machines supported by those
structures. As noted in Section 26.15.1, the VMXOFF instruction cannot be used
under the dual-monitor treatment and thus cannot be used by an SMM monitor.
The RSM instruction also cannot be used under the dual-monitor treatment. As noted
in Section 22.1.3, it causes a VM exit if executed in SMM in VMX non-root operation.
If executed in VMX root operation, it causes an invalid-opcode exception. SMM
monitor uses VM entries to return from SMM (see Section 26.15.4).
26.15.4 VM Entries that Return from SMM
The SMM monitor returns from SMM using a VM entry with the “entry to SMM”
VM-entry control clear. VM entries that return from SMM reverse the effects of an
SMM VM exit (see Section 26.15.2).
VM entries that return from SMM may differ from other VM entries in that they do not
necessarily enter VMX non-root operation. If the executive-VMCS pointer field in the
current VMCS contains the VMXON pointer, the logical processor remains in VMX root
operation after VM entry.
For differences between VM entries that return from SMM and other VM entries see
Sections 26.15.4.1 through 26.15.4.10.
26-30 Vol. 3B
SYSTEM MANAGEMENT MODE
26.15.4.1 Checks on the Executive-VMCS Pointer Field
VM entries that return from SMM perform the following checks on the executiveVMCS pointer field in the current VMCS:
•
•
Bits 11:0 must be 0.
•
The 32 bits located in memory referenced by the physical address in the pointer
must contain the processor’s VMCS revision identifier (see Section 21.2).
The pointer must not set any bits beyond the processor’s physical-address
width.1,2
The checks above are performed before the checks described in Section 26.15.4.2
and before any of the following checks:
•
If the “deactivate dual-monitor treatment” VM-entry control is 0, the launch state
of the executive VMCS (the VMCS referenced by the executive-VMCS pointer
field) must be launched (see Section 21.10.3).
•
If the “deactivate dual-monitor treatment” VM-entry control is 1, the executiveVMCS pointer field must contain the VMXON pointer (see Section 26.15.7).3
26.15.4.2 Checks on VM-Execution Control Fields
VM entries that return from SMM differ from other VM entries with regard to the
checks performed on the VM-execution control fields specified in Section 23.2.1.1.
They do not apply the checks to the current VMCS. Instead, VM-entry behavior
depends on whether the executive-VMCS pointer field contains the VMXON pointer:
•
If the executive-VMCS pointer field contains the VMXON pointer (the VM entry
remains in VMX root operation), the checks are not performed at all.
•
If the executive-VMCS pointer field does not contain the VMXON pointer (the
VM entry enters VMX non-root operation), the checks are performed on the
VM-execution control fields in the executive VMCS (the VMCS referenced by the
executive-VMCS pointer field in the current VMCS). These checks are performed
after checking the executive-VMCS pointer field itself (for proper alignment).
Other VM entries ensure that, if “activate VMX-preemption timer” VM-execution
control is 0, the “save VMX-preemption timer value” VM-exit control is also 0. This
check is not performed by VM entries that return from SMM.
1. Software can determine a processor’s physical-address width by executing CPUID with
80000008H in EAX. The physical-address width is returned in bits 7:0 of EAX.
2. If IA32_VMX_BASIC[48] is read as 1, this pointer must not set any bits in the range 63:32; see
Appendix G.1.
3. An SMM monitor can determine the VMXON pointer by reading the executive-VMCS pointer field
in the current VMCS after the SMM VM exit that activates the dual-monitor treatment.
Vol. 3B 26-31
SYSTEM MANAGEMENT MODE
26.15.4.3 Checks on VM-Entry Control Fields
VM entries that return from SMM differ from other VM entries with regard to the
checks performed on the VM-entry control fields specified in Section 23.2.1.3.
Specifically, if the executive-VMCS pointer field contains the VMXON pointer (the
VM entry remains in VMX root operation), the following must not all hold for the
VM-entry interruption-information field:
•
•
•
the valid bit (bit 31) in the VM-entry interruption-information field is 1
the interruption type (bits 10:8) is not 7 (other event); and
the vector (bits 7:0) is not 0 (pending MTF VM exit).
26.15.4.4 Checks on the Guest State Area
Section 23.3.1 specifies checks performed on fields in the guest-state area of the
VMCS. Some of these checks are conditioned on the settings of certain VM-execution
controls (e.g., “virtual NMIs” or “unrestricted guest”). VM entries that return from
SMM modify these checks based on whether the executive-VMCS pointer field
contains the VMXON pointer:1
•
If the executive-VMCS pointer field contains the VMXON pointer (the VM entry
remains in VMX root operation), the checks are performed as all relevant VMexecution controls were 0. (As a result, some checks may not be performed at
all.)
•
If the executive-VMCS pointer field does not contain the VMXON pointer (the
VM entry enters VMX non-root operation), this check is performed based on the
settings of the VM-execution controls in the executive VMCS (the VMCS
referenced by the executive-VMCS pointer field in the current VMCS).
For VM entries that return from SMM, the activity-state field must not indicate the
wait-for-SIPI state if the executive-VMCS pointer field contains the VMXON pointer
(the VM entry is to VMX root operation).
26.15.4.5 Loading Guest State
VM entries that return from SMM load the SMBASE register from the SMBASE field.
VM entries that return from SMM invalidate linear mappings and combined mappings
associated with all VPIDs. Combined mappings are invalidated for all EP4TA values
(EP4TA is the value of bits 51:12 of EPTP; see Section 25.3). (Ordinary VM entries
are required to perform such invalidation only for VPID 0000H and are not required
to do even that if the “enable VPID” VM-execution control is 1; see Section 23.3.2.5.)
1. An SMM monitor can determine the VMXON pointer by reading the executive-VMCS pointer field
in the current VMCS after the SMM VM exit that activates the dual-monitor treatment.
26-32 Vol. 3B
SYSTEM MANAGEMENT MODE
26.15.4.6 VMX-Preemption Timer
A VM entry that returns from SMM activates the VMX-preemption timer only if the
executive-VMCS pointer field does not contain the VMXON pointer (the VM entry
enters VMX non-root operation) and the “activate VMX-preemption timer” VM-entry
control is 1 in the executive VMCS (the VMCS referenced by the executive-VMCS
pointer field). In this case, VM entry starts the VMX-preemption timer with the value
in the VMX-preemption timer-value field in the current VMCS.
26.15.4.7 Updating the Current-VMCS and SMM-Transfer VMCS Pointers
Successful VM entries (returning from SMM) load the SMM-transfer VMCS pointer
with the current-VMCS pointer. Following this, they load the current-VMCS pointer
from a field in the current VMCS:
•
If the executive-VMCS pointer field contains the VMXON pointer (the VM entry
remains in VMX root operation), the current-VMCS pointer is loaded from the
VMCS-link pointer field.
•
If the executive-VMCS pointer field does not contain the VMXON pointer (the
VM entry enters VMX non-root operation), the current-VMCS pointer is loaded
with the value of the executive-VMCS pointer field.
If the VM entry successfully enters VMX non-root operation, the VM-execution
controls in effect after the VM entry are those from the new current VMCS. This
includes any structures external to the VMCS referenced by VM-execution control
fields.
The updating of these VMCS pointers occurs before event injection. Event injection is
determined, however, by the VM-entry control fields in the VMCS that was current
when the VM entry commenced.
26.15.4.8 VM Exits Induced by VM Entry
Section 23.5.1.2 describes how the event-delivery process invoked by event injection may lead to a VM exit. Section 23.6.3 to Section 23.6.7 describe other situations
that may cause a VM exit to occur immediately after a VM entry.
Whether these VM exits occur is determined by the VM-execution control fields in the
current VMCS. For VM entries that return from SMM, they can occur only if the executive-VMCS pointer field does not contain the VMXON pointer (the VM entry enters
VMX non-root operation).
In this case, determination is based on the VM-execution control fields in the VMCS
that is current after the VM entry. This is the VMCS referenced by the value of the
executive-VMCS pointer field at the time of the VM entry (see Section 26.15.4.7).
This VMCS also controls the delivery of such VM exits. Thus, VM exits induced by a
VM entry returning from SMM are to the executive monitor and not to the SMM
monitor.
Vol. 3B 26-33
SYSTEM MANAGEMENT MODE
26.15.4.9 SMI Blocking
VM entries that return from SMM determine the blocking of system-management
interrupts (SMIs) as follows:
•
If the “deactivate dual-monitor treatment” VM-entry control is 0, SMIs are
blocked after VM entry if and only if the bit 2 in the interruptibility-state field is 1.
•
If the “deactivate dual-monitor treatment” VM-entry control is 1, the blocking of
SMIs depends on whether the logical processor is in SMX operation:1
— If the logical processor is in SMX operation, SMIs are blocked after VM entry.
— If the logical processor is outside SMX operation, SMIs are unblocked after
VM entry.
VM entries that return from SMM and that do not deactivate the dual-monitor treatment may leave SMIs blocked. This feature exists to allow an SMM monitor to invoke
functionality outside of SMM without unblocking SMIs.
26.15.4.10 Failures of VM Entries That Return from SMM
Section 23.7 describes the treatment of VM entries that fail during or after loading
guest state. Such failures record information in the VM-exit information fields and
load processor state as would be done on a VM exit. The VMCS used is the one that
was current before the VM entry commenced. Control is thus transferred to the SMM
monitor and the logical processor remains in SMM.
26.15.5 Enabling the Dual-Monitor Treatment
Code and data for the SMM monitor reside in a region of SMRAM called the monitor
segment (MSEG). Code running in SMM determines the location of MSEG and establishes its content. This code is also responsible for enabling the dual-monitor treatment.
SMM code enables the dual-monitor treatment and determines the location of MSEG
by writing to IA32_SMM_MONITOR_CTL MSR (index 9BH). The MSR has the following
format:
•
Bit 0 is the register’s valid bit. The SMM monitor may be invoked using VMCALL
only if this bit is 1. Because VMCALL is used to activate the dual-monitor
treatment (see Section 26.15.6), the dual-monitor treatment cannot be
activated if the bit is 0. This bit is cleared when the logical processor is reset.
•
Bit 1 is reserved.
1. A logical processor is in SMX operation if GETSEC[SEXIT] has not been executed since the last
execution of GETSEC[SENTER]. A logical processor is outside SMX operation if GETSEC[SENTER]
has not been executed or if GETSEC[SEXIT] was executed after the last execution of GETSEC[SENTER]. See Chapter 6, “Safer Mode Extensions Reference,” in Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 2B.
26-34 Vol. 3B
SYSTEM MANAGEMENT MODE
•
Bit 2 determines whether executions of VMXOFF unblock SMIs under the default
treatment of SMIs and SMM. Executions of VMXOFF unblock SMIs unless bit 2 is
1 (the value of bit 0 is irrelevant). See Section 26.14.4.
Certain leaf functions of the GETSEC instruction clear this bit (see Chapter 6,
“Safer Mode Extensions Reference,” in Intel® 64 and IA-32 Architectures
Software Developer’s Manual, Volume 2B)
•
•
Bits 11:3 are reserved.
•
Bits 63:32 are reserved.
Bits 31:12 contain a value that, when shifted right 12 bits, is the physical address
of MSEG (the MSEG base address).
The following items detail use of this MSR:
•
The IA32_SMM_MONITOR_CTL MSR is supported only on processors that support
the dual-monitor treatment.1 On other processors, accesses to the MSR using
RDMSR or WRMSR generate a general-protection fault (#GP(0)).
•
A write to the IA32_SMM_MONITOR_CTL MSR using WRMSR generates a
general-protection fault (#GP(0)) if executed outside of SMM or if an attempt is
made to set any reserved bit. An attempt to write to IA32_SMM_MONITOR_CTL
MSR fails if made as part of a VM exit that does not end in SMM or part of a
VM entry that does not begin in SMM.
•
Reads from IA32_SMM_MONITOR_CTL MSR using RDMSR are allowed any time
RDMSR is allowed. The MSR may be read as part of any VM exit.
•
The dual-monitor treatment can be activated only if the valid bit in the MSR is set
to 1.
The 32 bytes located at the MSEG base address are called the MSEG header. The
format of the MSEG header is given in Table 26-10 (each field is 32 bits).
Table 26-10. Format of MSEG Header
Byte Offset
Field
0
MSEG-header revision identifier
4
SMM-monitor features
8
GDTR limit
12
GDTR base offset
16
CS selector
20
EIP offset
1. Software should consult the VMX capability MSR IA32_VMX_BASIC (see Appendix G.1) to determine whether the dual-monitor treatment is supported.
Vol. 3B 26-35
SYSTEM MANAGEMENT MODE
Table 26-10. Format of MSEG Header (Contd.)
Byte Offset
Field
24
ESP offset
28
CR3 offset
To ensure proper behavior in VMX operation, software should maintain the MSEG
header in writeback cacheable memory. Future implementations may allow or
require a different memory type.1 Software should consult the VMX capability MSR
IA32_VMX_BASIC (see Appendix G.1).
SMM code should enable the dual-monitor treatment (by setting the valid bit in
IA32_SMM_MONITOR_CTL MSR) only after establishing the content of the MSEG
header as follows:
•
Bytes 3:0 contain the MSEG revision identifier. Different processors may use
different MSEG revision identifiers. These identifiers enable software to avoid
using an MSEG header formatted for one processor on a processor that uses a
different format. Software can discover the MSEG revision identifier that a
processor uses by reading the VMX capability MSR IA32_VMX_MISC (see
Appendix G.6).
•
Bytes 7:4 contain the SMM-monitor features field. Bits 31:1 of this field are
reserved and must be zero. Bit 0 of the field is the IA-32e mode SMM feature
bit. It indicates whether the logical processor will be in IA-32e mode after the
SMM monitor is activated (see Section 26.15.6).
•
Bytes 31:8 contain fields that determine how processor state is loaded when the
SMM monitor is activated (see Section 26.15.6.4). SMM code should establish
these fields so that activating of the SMM monitor invokes the SMM monitor’s
initialization code.
26.15.6 Activating the Dual-Monitor Treatment
The dual-monitor treatment may be enabled by SMM code as described in Section
26.15.5. The dual-monitor treatment is activated only if it is enabled and only by the
executive monitor. The executive monitor activates the dual-monitor treatment by
executing VMCALL in VMX root operation.
When VMCALL activates the dual-monitor treatment, it causes an SMM VM exit.
Differences between this SMM VM exit and other SMM VM exits are discussed in
1. Alternatively, software may map the MSEG header with the UC memory type; this may be necessary, depending on how memory is organized. Doing so is strongly discouraged unless necessary
as it will cause the performance of transitions using those structures to suffer significantly. In
addition, the processor will continue to use the memory type reported in the VMX capability MSR
IA32_VMX_BASIC with exceptions noted in Appendix G.1.
26-36 Vol. 3B
SYSTEM MANAGEMENT MODE
Sections 26.15.6.1 through 26.15.6.5. See also “VMCALL—Call to VM Monitor” in
Chapter 6 of Intel® 64 and IA-32 Architectures Software Developer’s Manual,
Volume 2B.
26.15.6.1 Initial Checks
An execution of VMCALL attempts to activate the dual-monitor treatment if (1) the
processor supports the dual-monitor treatment;1 (2) the logical processor is in VMX
root operation; (3) the logical processor is outside SMM and the valid bit is set in the
IA32_SMM_MONITOR_CTL MSR; (4) the logical processor is not in virtual-8086
mode and not in compatibility mode; (5) CPL = 0; and (6) the dual-monitor treatment is not active.
The VMCS that manages SMM VM exit caused by this VMCALL is the current VMCS
established by the executive monitor. The VMCALL performs the following checks on
the current VMCS in the order indicated:
1. There must be a current VMCS pointer.
2. The launch state of the current VMCS must be clear.
3. The VM-exit control fields must be valid:
— Reserved bits in the VM-exit controls must be set properly. Software may
consult the VMX capability MSR IA32_VMX_EXIT_CTLS to determine the
proper settings (see Appendix G.4).
— The following checks are performed for the VM-exit MSR-store address if the
VM-exit MSR-store count field is non-zero:
•
The lower 4 bits of the VM-exit MSR-store address must be 0. The address
should not set any bits beyond the processor’s physical-address width.2
•
The address of the last byte in the VM-exit MSR-store area should not set
any bits beyond the processor’s physical-address width. The address of
this last byte is VM-exit MSR-store address + (MSR count * 16) – 1. (The
arithmetic used for the computation uses more bits than the processor’s
physical-address width.)
If IA32_VMX_BASIC[48] is read as 1, neither address should set any bits in
the range 63:32; see Appendix G.1.
If any of these checks fail, subsequent checks are skipped and VMCALL fails. If all
these checks succeed, the logical processor uses the IA32_SMM_MONITOR_CTL MSR
to determine the base address of MSEG. The following checks are performed in the
order indicated:
1. Software should consult the VMX capability MSR IA32_VMX_BASIC (see Appendix G.1) to determine whether the dual-monitor treatment is supported.
2. Software can determine a processor’s physical-address width by executing CPUID with
80000008H in EAX. The physical-address width is returned in bits 7:0 of EAX.
Vol. 3B 26-37
SYSTEM MANAGEMENT MODE
1. The logical processor reads the 32 bits at the base of MSEG and compares them
to the processor’s MSEG revision identifier.
2. The logical processor reads the SMM-monitor features field:
— Bit 0 of the field is the IA-32e mode SMM feature bit, and it indicates whether
the logical processor will be in IA-32e mode after the SMM monitor is
activated.
•
If the VMCALL is executed on a processor that does not support Intel 64
architecture, the IA-32e mode SMM feature bit must be 0.
•
If the VMCALL is executed in 64-bit mode, the IA-32e mode SMM feature
bit must be 1.
— Bits 31:1 of this field are currently reserved and must be zero.
If any of these checks fail, subsequent checks are skipped and the VMCALL fails.
26.15.6.2 MSEG Checking
SMM VM exits that activate the dual-monitor treatment check the following before
updating the current-VMCS pointer and the executive-VMCS pointer field (see
Section 26.15.2.2):
•
The 32 bits at the MSEG base address (used as a physical address) must contain
the processor’s MSEG revision identifier.
•
Bits 31:1 of the SMM-monitor features field in the MSEG header (see
Table 26-10) must be 0. Bit 0 of the field (the IA-32e mode SMM feature bit)
must be 0 if the processor does not support Intel 64 architecture.
If either of these checks fail, execution of VMCALL fails.
26.15.6.3 Updating the Current-VMCS and Executive-VMCS Pointers
Before performing the steps in Section 26.15.2.2, SMM VM exits that activate the
dual-monitor treatment begin by loading the SMM-transfer VMCS pointer with the
value of the current-VMCS pointer.
26.15.6.4 Loading Host State
The VMCS that is current during an SMM VM exit that activates the dual-monitor
treatment was established by the executive monitor. It does not contain the VM-exit
controls and host state required to initialize the SMM monitor. For this reason, such
SMM VM exits do not load processor state as described in Section 24.5. Instead,
state is set to fixed values or loaded based on the content of the MSEG header (see
Table 26-10):
•
CR0 is set to as follows:
— PG, NE, ET, MP, and PE are all set to 1.
26-38 Vol. 3B
SYSTEM MANAGEMENT MODE
— CD and NW are left unchanged.
— All other bits are cleared to 0.
•
CR3 is set as follows:
— Bits 63:32 are cleared on processors that supports IA-32e mode.
— Bits 31:12 are set to bits 31:12 of the sum of the MSEG base address and the
CR3-offset field in the MSEG header.
— Bits 11:5 and bits 2:0 are cleared (the corresponding bits in the CR3-offset
field in the MSEG header are ignored).
— Bits 4:3 are set to bits 4:3 of the CR3-offset field in the MSEG header.
•
CR4 is set as follows:
— MCE and PGE are cleared.
— PAE is set to the value of the IA-32e mode SMM feature bit.
— If the IA-32e mode SMM feature bit is clear, PSE is set to 1 if supported by the
processor; if the bit is set, PSE is cleared.
— All other bits are unchanged.
•
•
•
DR7 is set to 400H.
The IA32_DEBUGCTL MSR is cleared to 00000000_00000000H.
The registers CS, SS, DS, ES, FS, and GS are loaded as follows:
— All registers are usable.
— CS.selector is loaded from the corresponding fields in the MSEG header (the
high 16 bits are ignored), with bits 2:0 cleared to 0. If the result is 0000H,
CS.selector is set to 0008H.
— The selectors for SS, DS, ES, FS, and GS are set to CS.selector+0008H. If the
result is 0000H (if the CS selector was 0xFFF8), these selectors are instead
set to 0008H.
— The base addresses of all registers are cleared to zero.
— The segment limits for all registers are set to FFFFFFFFH.
— The AR bytes for the registers are set as follows:
•
CS.Type is set to 11 (execute/read, accessed, non-conforming code
segment).
•
For SS, DS, FS, and GS, the Type is set to 3 (read/write, accessed,
expand-up data segment).
•
•
•
The S bits for all registers are set to 1.
The DPL for each register is set to 0.
The P bits for all registers are set to 1.
Vol. 3B 26-39
SYSTEM MANAGEMENT MODE
•
On processors that support Intel 64 architecture, CS.L is loaded with the
value of the IA-32e mode SMM feature bit.
•
CS.D is loaded with the inverse of the value of the IA-32e mode SMM
feature bit.
•
•
For each of SS, DS, FS, and GS, the D/B bit is set to 1.
The G bits for all registers are set to 1.
•
LDTR is unusable. The LDTR selector is cleared to 0000H, and the register is
otherwise undefined (although the base address is always canonical)
•
GDTR.base is set to the sum of the MSEG base address and the GDTR base-offset
field in the MSEG header (bits 63:32 are always cleared on processors that
supports IA-32e mode). GDTR.limit is set to the corresponding field in the MSEG
header (the high 16 bits are ignored).
•
•
IDTR.base is unchanged. IDTR.limit is cleared to 0000H.
•
RSP is set to the sum of the MSEG base address and the value of the RSP-offset
field in the MSEG header (bits 63:32 are always cleared on logical processor that
supports IA-32e mode).
•
•
•
RFLAGS is cleared, except bit 1, which is always set.
RIP is set to the sum of the MSEG base address and the value of the RIP-offset
field in the MSEG header (bits 63:32 are always cleared on logical processors
that support IA-32e mode).
The logical processor is left in the active state.
Event blocking after the SMM VM exit is as follows:
— There is no blocking by STI or by MOV SS.
— There is blocking by non-maskable interrupts (NMIs) and by SMIs.
•
•
There are no pending debug exceptions after the SMM VM exit.
For processors that support IA-32e mode, the IA32_EFER MSR is modified so that
LME and LMA both contain the value of the IA-32e mode SMM feature bit.
If any of CR3[63:5], CR4.PAE, CR4.PSE, or IA32_EFER.LMA is changing, the TLBs are
updated so that, after VM exit, the logical processor does not use translations that
were cached before the transition. This is not necessary for changes that would not
affect paging due to the settings of other bits (for example, changes to CR4.PSE if
IA32_EFER.LMA was 1 before and after the transition).
26.15.6.5 Loading MSRs
The VM-exit MSR-load area is not used by SMM VM exits that activate the dualmonitor treatment. No MSRs are loaded from that area.
26-40 Vol. 3B
SYSTEM MANAGEMENT MODE
26.15.7 Deactivating the Dual-Monitor Treatment
An SMM monitor may deactivate the dual monitor treatment and return the
processor to default treatment of SMIs and SMM (see Section 26.14). It does this by
executing a VM entry with the “deactivate dual-monitor treatment” VM-entry control
set to 1.
As noted in Section 23.2.1.3 and Section 26.15.4.1, an attempt to deactivate the
dual-monitor treatment fails in the following situations: (1) the processor is not in
SMM; (2) the “entry to SMM” VM-entry control is 1; or (3) the executive-VMCS
pointer does not contain the VMXON pointer (the VM entry is to VMX non-root operation).
As noted in Section 26.15.4.9, VM entries that deactivate the dual-monitor treatment ignore the SMI bit in the interruptibility-state field of the guest-state area.
Instead, the blocking of SMIs following such a VM entry depends on whether the
logical processor is in SMX operation:1
•
If the logical processor is in SMX operation, SMIs are blocked after VM entry.
SMIs may later be unblocked by the VMXOFF instruction (see Section 26.14.4) or
by certain leaf functions of the GETSEC instruction (see Chapter 6, “Safer Mode
Extensions Reference,” in Intel® 64 and IA-32 Architectures Software
Developer’s Manual, Volume 2B).
•
If the logical processor is outside SMX operation, SMIs are unblocked after
VM entry.
26.16
SMI AND PROCESSOR EXTENDED STATE
MANAGEMENT
On processors that support processor extended states using XSAVE/XRSTOR (see
Chapter 13, “System Programming for Instruction Set Extensions and Processor
Extended States”), the processor does not save any XSAVE/XRSTOR related state on
an SMI. It is the responsibility of the SMM handler code to properly preserve the state
information (including CR4.OSXSAVE, XCR0, and possibly processor extended states
using XSAVE/XRSTOR). Therefore, the SMM handler must follow the rules described
in Chapter 13.
1. A logical processor is in SMX operation if GETSEC[SEXIT] has not been executed since the last
execution of GETSEC[SENTER]. A logical processor is outside SMX operation if GETSEC[SENTER]
has not been executed or if GETSEC[SEXIT] was executed after the last execution of GETSEC[SENTER]. See Chapter 6, “Safer Mode Extensions Reference,” in Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 2B.
Vol. 3B 26-41
SYSTEM MANAGEMENT MODE
26-42 Vol. 3B
CHAPTER 27
VIRTUAL-MACHINE MONITOR PROGRAMMING
CONSIDERATIONS
27.1
VMX SYSTEM PROGRAMMING OVERVIEW
The Virtual Machine Monitor (VMM) is a software class used to manage virtual
machines (VM). This chapter describes programming considerations for VMMs.
Each VM behaves like a complete physical machine and can run operating system
(OS) and applications. The VMM software layer runs at the most privileged level and
has complete ownership of the underlying system hardware. The VMM controls
creation of a VM, transfers control to a VM, and manages situations that can cause
transitions between the guest VMs and host VMM. The VMM allows the VMs to share
the underlying hardware and yet provides isolation between the VMs. The guest software executing in a VM is unaware of any transitions that might have occurred
between the VM and its host.
27.2
SUPPORTING PROCESSOR OPERATING MODES IN
GUEST ENVIRONMENTS
Typically, VMMs transfer control to a VM using VMX transitions referred to as VM
entries. The boundary conditions that define what a VM is allowed to execute in isolation are specified in a virtual-machine control structure (VMCS).
As noted in Section 20.8, processors may fix certain bits in CR0 and CR4 to specific
values and not support other values. The first processors to support VMX operation
require that CR0.PE and CR0.PG be 1 in VMX operation. Thus, a VM entry is allowed
only to guests with paging enabled that are in protected mode or in virtual-8086
mode. Guest execution in other processor operating modes need to be specially
handled by the VMM.
One example of such a condition is guest execution in real-mode. A VMM could
support guest real-mode execution using at least two approaches:
•
•
By using a fast instruction set emulator in the VMM.
By using the similarity between real-mode and virtual-8086 mode to support
real-mode guest execution in a virtual-8086 container. The virtual-8086
container may be implemented as a virtual-8086 container task within a monitor
that emulates real-mode guest state and instructions, or by running the guest VM
as the virtual-8086 container (by entering the guest with RFLAGS.VM1 set).
Attempts by real-mode code to access privileged state outside the virtual-8086
container would trap to the VMM and would also need to be emulated.
Vol. 3B 27-1
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
Another example of such a condition is guest execution in protected mode with
paging disabled. A VMM could support such guest execution by using “identity” page
tables to emulate unpaged protected mode.
27.2.1
Using Unrestricted Guest Mode
Processors which support the “unrestricted guest” VM-execution control allow VM
software to run in real-address mode and unpaged protected mode. Since these
modes do not use paging, VMM software must virtualize guest memory using EPT.
Special notes for 64-bit VMM software using the 1-setting of the “unrestricted guest”
VM-execution control:
•
It is recommended that 64-bit VMM software use the 1-settings of the "load
IA32_EFER" VM entry control and the "save IA32_EFER" VM-exit control. If VM
entry is establishing CR0.PG=0 and if the "IA-32e mode guest" and "load
IA32_EFER" VM entry controls are both 0, VM entry leaves IA32_EFER.LME
unmodified (i.e., the host value will persist in the guest).
•
It is not necessary for VMM software to track guest transitions into and out of IA32e mode for the purpose of maintaining the correct setting of the "IA-32e mode
guest" VM entry control. This is because VM exits on processors supporting the
1-setting of the "unrestricted guest" VM-execution control save the (guest) value
of IA32_EFER.LMA into the "IA-32e mode guest" VM entry control.
27.3
MANAGING VMCS REGIONS AND POINTERS
A VMM must observe necessary procedures when working with a VMCS, the associated VMCS pointer, and the VMCS region. It must also not assume the state of persistency for VMCS regions in memory or cache.
Before entering VMX operation, the host VMM allocates a VMXON region. A VMM can
host several virtual machines and have many VMCSs active under its management.
A unique VMCS region is required for each virtual machine; a VMXON region is
required for the VMM itself.
A VMM determines the VMCS region size by reading IA32_VMX_BASIC MSR; it
creates VMCS regions of this size using a 4-KByte-aligned area of physical memory.
Each VMCS region needs to be initialized with a VMCS revision identifier (at byte
offset 0) identical to the revision reported by the processor in the VMX capability
MSR.
1. This chapter uses the notation RAX, RIP, RSP, RFLAGS, etc. for processor registers because most
processors that support VMX operation also support Intel 64 architecture. For processors that do
not support Intel 64 architecture, this notation refers to the 32-bit forms of those registers
(EAX, EIP, ESP, EFLAGS, etc.).
27-2 Vol. 3B
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
NOTE
Software must not read or write directly to the VMCS data region as
the format is not architecturally defined. Consequently, Intel
recommends that the VMM remove any linear-address mappings to
VMCS regions before loading.
System software does not need to do special preparation to the VMXON region before
entering into VMX operation. The address of the VMXON region for the VMM is
provided as an operand to VMXON instruction. Once in VMX root operation, the VMM
needs to prepare data fields in the VMCS that control the execution of a VM upon a
VM entry. The VMM can make a VMCS the current VMCS by using the VMPTRLD
instruction. VMCS data fields must be read or written only through VMREAD and
VMWRITE commands respectively.
Every component of the VMCS is identified by a 32-bit encoding that is provided as
an operand to VMREAD and VMWRITE. Appendix H provides the encodings. A VMM
must properly initialize all fields in a VMCS before using the current VMCS for VM
entry.
A VMCS is referred to as a controlling VMCS if it is the current VMCS on a logical
processor in VMX non-root operation. A current VMCS for controlling a logical
processor in VMX non-root operation may be referred to as a working VMCS if the
logical processor is not in VMX non-root operation. The relationship of active, current
(i.e. working) and controlling VMCS during VMX operation is shown in Figure 27-1.
NOTE
As noted in Section 21.1, the processor may optimize VMX operation
by maintaining the state of an active VMCS (one for which VMPTRLD
has been executed) on the processor. Before relinquishing control to
other system software that may, without informing the VMM, remove
power from the processor (e.g., for transitions to S3 or S4) or leave
VMX operation, a VMM must VMCLEAR all active VMCSs. This ensures
that all VMCS data cached by the processor are flushed to memory
and that no other software can corrupt the current VMM’s VMCS data.
It is also recommended that the VMM execute VMXOFF after such
executions of VMCLEAR.
The VMX capability MSR IA32_VMX_BASIC reports the memory type used by the
processor for accessing a VMCS or any data structures referenced through pointers in
the VMCS. Software must maintain the VMCS structures in cache-coherent memory.
Software must always map the regions hosting the I/O bitmaps, MSR bitmaps, VMexit MSR-store area, VM-exit MSR-load area, and VM-entry MSR-load area to the
write-back (WB) memory type. Mapping these regions to uncacheable (UC) memory
type is supported, but strongly discouraged due to negative impact on performance.
Vol. 3B 27-3
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
(a) VMX Operation and VMX Transitions
VM Entry
VM Entry
VM Entry
VM Entry
VMXOFF
Processor
Operation
VMXON
VM Exit
VM Exit
VM Exit
VM Exit
Legend:
VMX Root
Operation
Outside
VMX
Operation
VMX
Non-Root
Operation
(b) State of VMCS and VMX Operation
VMLAUNCH
VMPTRLD B
VMRESUME
VMCLEAR B
VM Exit
VM Exit
VMCS B
VMCS A
VMPTRLD A
VMPTRLD A
VM Exit
VMLAUNCH
Legend:
Inactive
VMCS
Current VMCS
(working)
Active VMCS
(not current)
VM Exit
VMRESUME
VMCLEAR A
Current VMCS
(controlling)
Figure 27-1. VMX Transitions and States of VMCS in a Logical Processor
27.4
USING VMX INSTRUCTIONS
VMX instructions are allowed only in VMX root operation. An attempt to execute a
VMX instruction in VMX non-root operation causes a VM exit.
27-4 Vol. 3B
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
Processors perform various checks while executing any VMX instruction. They follow
well-defined error handling on failures. VMX instruction execution failures detected
before loading of a guest state are handled by the processor as follows:
•
If the working-VMCS pointer is not valid, the instruction fails by setting
RFLAGS.CF to 1.
•
If the working-VMCS pointer is valid, RFLAGS.ZF is set to 1 and the proper errorcode is saved in the VM-instruction error field of the working-VMCS.
Software is required to check RFLAGS.CF and RFLAGS.ZF to determine the success or
failure of VMX instruction executions.
The following items provide details regarding use of the VM-entry instructions
(VMLAUNCH and VMRESUME):
•
If the working-VMCS pointer is valid, the state of the working VMCS may cause
the VM-entry instruction to fail. RFLAGS.ZF is set to 1 and one of the following
values is saved in the VM-instruction error field:
— 4: VMLAUNCH with non-clear VMCS.
If this error occurs, software can avoid the error by executing VMRESUME.
— 5: VMRESUME with non-launched VMCS.
If this error occurs, software can avoid the error by executing VMLAUNCH.
— 6: VMRESUME after VMXOFF.1
If this error occurs, software can avoid the error by executing the following
sequence of instructions:
VMPTRST working-VMCS pointer
VMCLEAR working-VMCS pointer
VMPTRLD working-VMCS pointer
VMLAUNCH
(VMPTRST may not be necessary is software already knows the workingVMCS pointer.)
•
If none of the above errors occur, the processor checks on the VMX controls and
host-state area. If any of these checks fail, the VM-entry instruction fails.
RFLAGS.ZF is set to 1 and either 7 (VM entry with invalid control field(s)) or 8
(VM entry with invalid host-state field(s)) is saved in the VM-instruction error
field.
•
After a VM-entry instruction (VMRESUME or VMLAUNCH) successfully completes
the general checks and checks on VMX controls and the host-state area (see
Section 23.2), any errors encountered while loading of guest-state (due to bad
guest-state or bad MSR loading) causes the processor to load state from the
host-state area of the working VMCS as if a VM exit had occurred (see Section
27.7).
1. Earlier versions of this manual described this error as “VMRESUME with a corrupted VMCS”.
Vol. 3B 27-5
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
This failure behavior differs from that of VM exits in that no guest-state is saved
to the guest-state area. A VMM can detect its VM-exit handler was invoked by
such a failure by checking bit 31 (for 1) in the exit reason field of the working
VMCS and further identify the failure by using the exit qualification field.
See Chapter 23 for more details about the VM-entry instructions.
27.5
VMM SETUP & TEAR DOWN
VMMs need to ensure that the processor is running in protected mode with paging
before entering VMX operation. The following list describes the minimal steps
required to enter VMX root operation with a VMM running at CPL = 0.
•
•
Check VMX support in processor using CPUID.
•
Create a VMXON region in non-pageable memory of a size specified by
IA32_VMX_BASIC MSR and aligned to a 4-KByte boundary. Software should read
the capability MSRs to determine width of the physical addresses that may be
used for the VMXON region and ensure the entire VMXON region can be
addressed by addresses with that width. Also, software must ensure that the
VMXON region is hosted in cache-coherent memory.
•
Initialize the version identifier in the VMXON region (the first 32 bits) with the
VMCS revision identifier reported by capability MSRs.
•
Ensure the current processor operating mode meets the required CR0 fixed bits
(CR0.PE = 1, CR0.PG = 1). Other required CR0 fixed bits can be detected
through the IA32_VMX_CR0_FIXED0 and IA32_VMX_CR0_FIXED1 MSRs.
•
Enable VMX operation by setting CR4.VMXE = 1. Ensure the resultant CR4 value
supports all the CR4 fixed bits reported in the IA32_VMX_CR4_FIXED0 and
IA32_VMX_CR4_FIXED1 MSRs.
•
Ensure that the IA32_FEATURE_CONTROL MSR (MSR index 3AH) has been
properly programmed and that its lock bit is set (Bit 0 = 1). This MSR is generally
configured by the BIOS using WRMSR.
•
Execute VMXON with the physical address of the VMXON region as the operand.
Check successful execution of VMXON by checking if RFLAGS.CF = 0.
Determine the VMX capabilities supported by the processor through the VMX
capability MSRs. See Section 27.5.1 and Appendix G.
Upon successful execution of the steps above, the processor is in VMX root operation.
A VMM executing in VMX root operation and CPL = 0 leaves VMX operation by
executing VMXOFF and verifies successful execution by checking if RFLAGS.CF = 0
and RFLAGS.ZF = 0.
If an SMM monitor has been configured to service SMIs while in VMX operation (see
Section 26.15), the SMM monitor needs to be torn down before the executive
monitor can leave VMX operation (see Section 26.15.7). VMXOFF fails for the execu-
27-6 Vol. 3B
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
tive monitor (a VMM that entered VMX operation by way of issuing VMXON) if SMM
monitor is configured.
27.5.1
Algorithms for Determining VMX Capabilities
As noted earlier, a VMM should determine the VMX capabilities supported by the
processor by reading the VMX capability MSRs. The architecture for these MSRs is
detailed in Appendix G.
As noted in Chapter 21, “Virtual-Machine Control Structures”, certain VMX controls
are reserved and must be set to a specific value (0 or 1) determined by the processor.
The specific value to which a reserved control must be set is its default setting.
Most controls have a default setting of 0; Appendix G.2 identifies those controls that
have a default setting of 1. The term default1 describes the class of controls whose
default setting is 1. The are controls in this class from the pin-based VM-execution
controls, the primary processor-based VM-execution controls, the VM-exit controls,
and the VM-entry controls. There are no secondary processor-based VM-execution
controls in the default1 class.
Future processors may define new functionality for one or more reserved controls.
Such processors would allow each newly defined control to be set either to 0 or to 1.
Software that does not desire a control’s new functionality should set the control to
its default setting.
The capability MSRs IA32_VMX_PINBASED_CTLS, IA32_VMX_PROCBASED_CTLS,
IA32_VMX_EXIT_CTLS, and IA32_VMX_ENTRY_CTLS report, respectively, on the
allowed settings of most of the pin-based VM-execution controls, the primary
processor-based VM-execution controls, the VM-exit controls, and the VM-entry
controls. However, they will always report that any control in the default1 class must
be 1. If a logical processor allows any control in the default1 class to be 0, it indicates
this fact by returning 1 for the value of bit 55 of the IA32_VMX_BASIC MSR. If this bit
is 1, the logical processor supports the capability MSRs
IA32_VMX_TRUE_PINBASED_CTLS, IA32_VMX_TRUE_PROCBASED_CTLS,
IA32_VMX_TRUE_EXIT_CTLS, and IA32_VMX_TRUE_ENTRY_CTLS. These capability
MSRs report, respectively, on the allowed settings of all of the pin-based VM-execution controls, the primary processor-based VM-execution controls, the VM-exit
controls, and the VM-entry controls.
Software may use one of the following high-level algorithms to determine the correct
default control settings:1
1. The following algorithm does not use the details given in Appendix G.2:
a. Ignore bit 55 of the IA32_VMX_BASIC MSR.
1. These algorithms apply only to the pin-based VM-execution controls, the primary processorbased VM-execution controls, the VM-exit controls, and the VM-entry controls. Because there are
no secondary processor-based VM-execution controls in the default1 class, a VMM can always
set to 0 any such control whose meaning is unknown to it.
Vol. 3B 27-7
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
b. Using RDMSR, read the VMX capability MSRs IA32_VMX_PINBASED_CTLS,
IA32_VMX_PROCBASED_CTLS, IA32_VMX_EXIT_CTLS, and
IA32_VMX_ENTRY_CTLS.
c.
Set the VMX controls as follows:
i)
If the relevant VMX capability MSR reports that a control has a single
setting, use that setting.
ii) If (1) the relevant VMX capability MSR reports that a control can be set to
0 or 1; and (2) the control’s meaning is known to the VMM; then set the
control based on functionality desired.
iii) If (1) the relevant VMX capability MSR reports that a control can be set to
0 or 1; and (2) the control’s meaning is not known to the VMM; then set
the control to 0.
A VMM using this algorithm will set to 1 all controls in the default1 class (in
step (c)(i)). It will operate correctly even on processors that allow some
controls in the default1 class to be 0. However, such a VMM will not be able to
use the new features enabled by the 0-setting of such controls. For that reason,
this algorithm is not recommended.
2. The following algorithm uses the details given in Appendix G.2. This algorithm
requires software to know the identity of the controls in the default1 class:
a. Using RDMSR, read the IA32_VMX_BASIC MSR.
b. Use bit 55 of that MSR as follows:
i)
If bit 55 is 0, use RDMSR to read the VMX capability MSRs
IA32_VMX_PINBASED_CTLS, IA32_VMX_PROCBASED_CTLS,
IA32_VMX_EXIT_CTLS, and IA32_VMX_ENTRY_CTLS.
ii) If bit 55 is 1, use RDMSR to read the VMX capability MSRs
IA32_VMX_TRUE_PINBASED_CTLS,
IA32_VMX_TRUE_PROCBASED_CTLS, IA32_VMX_TRUE_EXIT_CTLS, and
IA32_VMX_TRUE_ENTRY_CTLS.
c.
Set the VMX controls as follows:
i)
If the relevant VMX capability MSR reports that a control has a single
setting, use that setting.
ii) If (1) the relevant VMX capability MSR reports that a control can be set to
0 or 1; and (2) the control’s meaning is known to the VMM; then set the
control based on functionality desired.
iii) If (1) the relevant VMX capability MSR reports that a control can be set to
0 or 1; (2) the control’s meaning is not known to the VMM; and (3) the
control is not in the default1 class; then set the control to 0.
iv) If (1) the relevant VMX capability MSR reports that a control can be set to
0 or 1; (2) the control’s meaning is not known to the VMM; and (3) the
control is in the default1 class; then set the control to 1.
27-8 Vol. 3B
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
A VMM using this algorithm will set to 1 all controls in default1 class whose
meaning it does not know (either in step (c)(i) or step (c)(iv)). It will operate
correctly even on processors that allow some controls in the default1 class to be
0. Unlike a VMM using Algorithm 1, a VMM using Algorithm 2 will be able to use
the new features enabled by the 0-setting of such controls.
3. The following algorithm uses the details given in Appendix G.2. This algorithm
does not require software to know the identity of the controls in the default1
class:
a. Using RDMSR, read the VMX capability MSRs IA32_VMX_BASIC,
IA32_VMX_PINBASED_CTLS, IA32_VMX_PROCBASED_CTLS,
IA32_VMX_EXIT_CTLS, and IA32_VMX_ENTRY_CTLS.
b. If bit 55 of the IA32_VMX_BASIC MSR is 0, set the VMX controls as follows:
i)
If the relevant VMX capability MSR reports that a control has a single
setting, use that setting.
ii) If (1) the relevant VMX capability MSR reports that a control can be set to
0 or 1; and (2) the control’s meaning is known to the VMM; then set the
control based on functionality desired.
iii) If (1) the relevant VMX capability MSR reports that a control can be set to
0 or 1; and (2) the control’s meaning is not known to the VMM; then set
the control to 0.
c.
If bit 55 of the IA32_VMX_BASIC MSR is 1, use RDMSR to read the VMX
capability MSRs IA32_VMX_TRUE_PINBASED_CTLS,
IA32_VMX_TRUE_PROCBASED_CTLS, IA32_VMX_TRUE_EXIT_CTLS, and
IA32_VMX_TRUE_ENTRY_CTLS. Set the VMX controls as follows:
i)
If the relevant VMX capability MSR just read reports that a control has a
single setting, use that setting.
ii) If (1) the relevant VMX capability MSR just read reports that a control can
be set to 0 or 1; and (2) the control’s meaning is known to the VMM; then
set the control based on functionality desired.
iii) If (1) the relevant VMX capability MSR just read reports that a control can
be set to 0 or 1; (2) the control’s meaning is not known to the VMM; and
(3) the relevant VMX capability MSR as read in step (a) reports that a
control can be set to 0; then set the control to 0.
iv) If (1) the relevant VMX capability MSR just read reports that a control can
be set to 0 or 1; (2) the control’s meaning is not known to the VMM; and
(3) the relevant VMX capability MSR as read in step (a) reports that a
control must be 1; then set the control to 1.
A VMM using this algorithm will set to 1 all controls in the default1 class whose
meaning it does not know (in step (b)(i), step (c)(i), or step (c)(iv)). It will
operate correctly even on processors that allow some controls in the default1
class to be 0. Unlike a VMM using Algorithm 1, a VMM using Algorithm 3 will be
able to use the new features enabled by the 0-setting of such controls. Unlike a
Vol. 3B 27-9
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
VMM using Algorithm 2, a VMM using Algorithm 3 need not know the identities
of the controls in the default1 class.
27.6
PREPARATION AND LAUNCHING A VIRTUAL
MACHINE
The following list describes the minimal steps required by the VMM to set up and
launch a guest VM.
•
Create a VMCS region in non-pageable memory of size specified by the VMX
capability MSR IA32_VMX_BASIC and aligned to 4-KBytes. Software should read
the capability MSRs to determine width of the physical addresses that may be
used for a VMCS region and ensure the entire VMCS region can be addressed by
addresses with that width. The term “guest-VMCS address” refers to the physical
address of the new VMCS region for the following steps.
•
Initialize the version identifier in the VMCS (first 32 bits) with the VMCS revision
identifier reported by the VMX capability MSR IA32_VMX_BASIC.
•
Execute the VMCLEAR instruction by supplying the guest-VMCS address. This will
initialize the new VMCS region in memory and set the launch state of the VMCS
to “clear”. This action also invalidates the working-VMCS pointer register to
FFFFFFFF_FFFFFFFFH. Software should verify successful execution of VMCLEAR
by checking if RFLAGS.CF = 0 and RFLAGS.ZF = 0.
•
Execute the VMPTRLD instruction by supplying the guest-VMCS address. This
initializes the working-VMCS pointer with the new VMCS region’s physical
address.
•
Issue a sequence of VMWRITEs to initialize various host-state area fields in the
working VMCS. The initialization sets up the context and entry-points to the VMM
upon subsequent VM exits from the guest. Host-state fields include control
registers (CR0, CR3 and CR4), selector fields for the segment registers (CS, SS,
DS, ES, FS, GS and TR), and base-address fields (for FS, GS, TR, GDTR and IDTR;
RSP, RIP and the MSRs that control fast system calls).
Chapter 22 describes the host-state consistency checking done by the processor
for VM entries. The VMM is required to set up host-state that comply with these
consistency checks. For example, VMX requires the host-area to have a task
register (TR) selector with TI and RPL fields set to 0 and pointing to a valid TSS.
•
Use VMWRITEs to set up the various VM-exit control fields, VM-entry control
fields, and VM-execution control fields in the VMCS. Care should be taken to
make sure the settings of individual fields match the allowed 0 and 1 settings for
the respective controls as reported by the VMX capability MSRs (see Appendix G).
Any settings inconsistent with the settings reported by the capability MSRs will
cause VM entries to fail.
•
Use VMWRITE to initialize various guest-state area fields in the working VMCS.
This sets up the context and entry-point for guest execution upon VM entry.
27-10 Vol. 3B
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
Chapter 22 describes the guest-state loading and checking done by the processor
for VM entries to protected and virtual-8086 guest execution.
•
The VMM is required to set up guest-state that complies with these consistency
checks:
— If the VMM design requires the initial VM launch to cause guest software
(typically the guest virtual BIOS) execution from the guest’s reset vector, it
may need to initialize the guest execution state to reflect the state of a
physical processor at power-on reset (described in Chapter 9, Intel® 64 and
IA-32 Architectures Software Developer’s Manual, Volume 3A).
— The VMM may need to initialize additional guest execution state that is not
captured in the VMCS guest-state area by loading them directly on the
respective processor registers. Examples include general purpose registers,
the CR2 control register, debug registers, floating point registers and so forth.
VMM may support lazy loading of FPU, MMX, SSE, and SSE2 states with
CR0.TS = 1 (described in Intel® 64 and IA-32 Architectures Software
Developer’s Manual, Volume 3A).
•
Execute VMLAUNCH to launch the guest VM. If VMLAUNCH fails due to any
consistency checks before guest-state loading, RFLAGS.CF or RFLAGS.ZF will be
set and the VM-instruction error field (see Section 21.9.5) will contain the errorcode. If guest-state consistency checks fail upon guest-state loading, the
processor loads state from the host-state area as if a VM exit had occurred (see
Section 27.6).
VMLAUNCH updates the controlling-VMCS pointer with the working-VMCS pointer
and saves the old value of controlling-VMCS as the parent pointer. In addition, the
launch state of the guest VMCS is changed to “launched” from “clear”. Any
programmed exit conditions will cause the guest to VM exit to the VMM. The VMM
should execute VMRESUME instruction for subsequent VM entries to guests in a
“launched” state.
27.7
HANDLING OF VM EXITS
This section provides examples of software steps involved in a VMM’s handling of VMexit conditions:
•
Determine the exit reason through a VMREAD of the exit-reason field in the
working-VMCS. Appendix I describes exit reasons and their encodings.
•
VMREAD the exit-qualification from the VMCS if the exit-reason field provides a
valid qualification. The exit-qualification field provides additional details on the
VM-exit condition. For example, in case of page faults, the exit-qualification field
provides the guest linear address that caused the page fault.
•
Depending on the exit reason, fetch other relevant fields from the VMCS.
Appendix I lists the various exit reasons.
Vol. 3B 27-11
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
•
Handle the VM-exit condition appropriately in the VMM. This may involve the
VMM emulating one or more guest instructions, programming the underlying
host hardware resources, and then re-entering the VM to continue execution.
27.7.1
Handling VM Exits Due to Exceptions
As noted in Section 22.3, an exception causes a VM exit if the bit corresponding to
the exception’s vector is set in the exception bitmap. (For page faults, the error code
also determines whether a VM exit occurs.) This section provide some guidelines of
how a VMM might handle such exceptions.
Exceptions result when a logical processor encounters an unusual condition that software may not have expected. When guest software encounters an exception, it may
be the case that the condition was caused by the guest software. For example, a
guest application may attempt to access a page that is restricted to supervisor
access. Alternatively, the condition causing the exception may have been established
by the VMM. For example, a guest OS may attempt to access a page that the VMM
has chosen to make not present.
When the condition causing an exception was established by guest software, the
VMM may choose to reflect the exception to guest software. When the condition was
established by the VMM itself, the VMM may choose to resume guest software after
removing the condition.
27.7.1.1
Reflecting Exceptions to Guest Software
If the VMM determines that a VM exit was caused by an exception due to a condition
established by guest software, it may reflect that exception to guest software. The
VMM would cause the exception to be delivered to guest software, where it can be
handled as it would be if the guest were running on a physical machine. This section
describes how that may be done.
In general, the VMM can deliver the exception to guest software using VM-entry
event injection as described in Section 23.5. The VMM can copy (using VMREAD and
VMWRITE) the contents of the VM-exit interruption-information field (which is valid,
since the VM exit was caused by an exception) to the VM-entry interruption-information field (which, if valid, will cause the exception to be delivered as part of the next
VM entry). The VMM would also copy the contents of the VM-exit interruption errorcode field to the VM-entry exception error-code field; this need not be done if bit 11
(error code valid) is clear in the VM-exit interruption-information field. After this, the
VMM can execute VMRESUME.
The following items provide details that may qualify the general approach:
•
Care should be taken to ensure that reserved bits 30:12 in the VM-entry interruption-information field are 0. In particular, some VM exits may set bit 12 in the
VM-exit interruption-information field to indicate NMI unblocking due to IRET. If
this bit is copied as 1 into the VM-entry interruption-information field, the next
VM entry will fail because that bit should be 0.
27-12 Vol. 3B
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
•
Bit 31 (valid) of the IDT-vectoring information field indicates, if set, that the
exception causing the VM exit occurred while another event was being delivered
to guest software. If this is the case, it may not be appropriate simply to reflect
that exception to guest software. To provide proper virtualization of the exception
architecture, a VMM should handle nested events as a physical processor would.
Processor handling is described in Chapter 6, “Interrupt 8—Double Fault
Exception (#DF)” in Intel® 64 and IA-32 Architectures Software Developer’s
Manual, Volume 3A.
— The VMM should reflect the exception causing the VM exit to guest software
in any of the following cases:
•
The value of bits 10:8 (interruption type) of the IDT-vectoring
information field is anything other than 3 (hardware exception).
•
The value of bits 7:0 (vector) of the IDT-vectoring information field
indicates a benign exception (1, 2, 3, 4, 5, 6, 7, 9, 16, 17, 18, or 19).
•
The value of bits 7:0 (vector) of the VM-exit interruption-information field
indicates a benign exception.
•
The value of bits 7:0 of the IDT-vectoring information field indicates a
contributory exception (0, 10, 11, 12, or 13) and the value of bits 7:0 of
the VM-exit interruption-information field indicates a page fault (14).
— If the value of bits 10:8 of the IDT-vectoring information field is 3 (hardware
exception), the VMM should reflect a double-fault exception to guest software
in any of the following cases:
•
The value of bits 7:0 of the IDT-vectoring information field and the value
of bits 7:0 of the VM-exit interruption-information field each indicates a
contributory exception.
•
The value of bits 7:0 of the IDT-vectoring information field indicates a
page fault and the value of bits 7:0 of the VM-exit interruptioninformation field indicates either a contributory exception or a page fault.
A VMM can reflect a double-fault exception to guest software by setting the
VM-entry interruption-information and VM-entry exception error-code fields
as follows:
•
Set bits 7:0 (vector) of the VM-entry interruption-information field to 8
(#DF).
•
Set bits 10:8 (interruption type) of the VM-entry interruption-information
field to 3 (hardware exception).
•
Set bit 11 (deliver error code) of the VM-entry interruption-information
field to 1.
•
•
•
Clear bits 30:12 (reserved) of VM-entry interruption-information field.
Set bit 31 (valid) of VM-entry interruption-information field.
Set the VM-entry exception error-code field to zero.
Vol. 3B 27-13
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
— If the value of bits 10:8 of the IDT-vectoring information field is 3 (hardware
exception) and the value of bits 7:0 is 8 (#DF), guest software would have
encountered a triple fault. Event injection should not be used in this case. The
VMM may choose to terminate the guest, or it might choose to enter the
guest in the shutdown activity state.
27.7.1.2
Resuming Guest Software after Handling an Exception
If the VMM determines that a VM exit was caused by an exception due to a condition
established by the VMM itself, it may choose to resume guest software after
removing the condition. The approach for removing the condition may be specific to
the VMM’s software architecture. and algorithms This section describes how guest
software may be resumed after removing the condition.
In general, the VMM can resume guest software simply by executing VMRESUME. The
following items provide details of cases that may require special handling:
•
If the “NMI exiting” VM-execution control is 0, bit 12 of the VM-exit interruptioninformation field indicates that the VM exit was due to a fault encountered during
an execution of the IRET instruction that unblocked non-maskable interrupts
(NMIs). In particular, it provides this indication if the following are both true:
— Bit 31 (valid) in the IDT-vectoring information field is 0.
— The value of bits 7:0 (vector) of the VM-exit interruption-information field is
not 8 (the VM exit is not due to a double-fault exception).
If both are true and bit 12 of the VM-exit interruption-information field is 1, NMIs
were blocked before guest software executed the IRET instruction that caused
the fault that caused the VM exit. The VMM should set bit 3 (blocking by NMI) in
the interruptibility-state field (using VMREAD and VMWRITE) before resuming
guest software.
•
If the “virtual NMIs” VM-execution control is 1, bit 12 of the VM-exit interruptioninformation field indicates that the VM exit was due to a fault encountered during
an execution of the IRET instruction that removed virtual-NMI blocking. In
particular, it provides this indication if the following are both true:
— Bit 31 (valid) in the IDT-vectoring information field is 0.
— The value of bits 7:0 (vector) of the VM-exit interruption-information field is
not 8 (the VM exit is not due to a double-fault exception).
If both are true and bit 12 of the VM-exit interruption-information field is 1, there
was virtual-NMI blocking before guest software executed the IRET instruction
that caused the fault that caused the VM exit. The VMM should set bit 3 (blocking
by NMI) in the interruptibility-state field (using VMREAD and VMWRITE) before
resuming guest software.
•
Bit 31 (valid) of the IDT-vectoring information field indicates, if set, that the
exception causing the VM exit occurred while another event was being delivered
to guest software. The VMM should ensure that the other event is delivered when
27-14 Vol. 3B
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
guest software is resumed. It can do so using the VM-entry event injection
described in Section 23.5 and detailed in the following paragraphs:
— The VMM can copy (using VMREAD and VMWRITE) the contents of the IDTvectoring information field (which is presumed valid) to the VM-entry interruption-information field (which, if valid, will cause the exception to be
delivered as part of the next VM entry).
•
The VMM should ensure that reserved bits 30:12 in the VM-entry interruption-information field are 0. In particular, the value of bit 12 in the IDTvectoring information field is undefined after all VM exits. If this bit is
copied as 1 into the VM-entry interruption-information field, the next
VM entry will fail because the bit should be 0.
•
If the “virtual NMIs” VM-execution control is 1 and the value of bits 10:8
(interruption type) in the IDT-vectoring information field is 2 (indicating
NMI), the VM exit occurred during delivery of an NMI that had been
injected as part of the previous VM entry. In this case, bit 3 (blocking by
NMI) will be 1 in the interruptibility-state field in the VMCS. The VMM
should clear this bit; otherwise, the next VM entry will fail (see Section
23.3.1.5).
— The VMM can also copy the contents of the IDT-vectoring error-code field to
the VM-entry exception error-code field. This need not be done if bit 11 (error
code valid) is clear in the IDT-vectoring information field.
— The VMM can also copy the contents of the VM-exit instruction-length field to
the VM-entry instruction-length field. This need be done only if bits 10:8
(interruption type) in the IDT-vectoring information field indicate either
software interrupt, privileged software exception, or software exception.
27.8
MULTI-PROCESSOR CONSIDERATIONS
The most common VMM design will be the symmetric VMM. This type of VMM runs the
same VMM binary on all logical processors. Like a symmetric operating system, the
symmetric VMM is written to ensure all critical data is updated by only one processor
at a time, IO devices are accessed sequentially, and so forth. Asymmetric VMM
designs are possible. For example, an asymmetric VMM may run its scheduler on one
processor and run just enough of the VMM on other processors to allow the correct
execution of guest VMs. The remainder of this section focuses on the multi-processor
considerations for a symmetric VMM.
A symmetric VMM design does not preclude asymmetry in its operations. For
example, a symmetric VMM can support asymmetric allocation of logical processor
resources to guests. Multiple logical processors can be brought into a single guest
environment to support an MP-aware guest OS. Because an active VMCS can not
control more than one logical processor simultaneously, a symmetric VMM must
make copies of its VMCS to control the VM allocated to support an MP-aware guest
Vol. 3B 27-15
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
OS. Care must be taken when accessing data structures shared between these
VMCSs. See Section 27.8.4.
Although it may be easier to develop a VMM that assumes a fully-symmetric view of
hardware capabilities (with all processors supporting the same processor feature
sets, including the same revision of VMX), there are advantages in developing a VMM
that comprehends different levels of VMX capability (reported by VMX capability
MSRs). One possible advantage of such an approach could be that an existing software installation (VMM and guest software stack) could continue to run without
requiring software upgrades to the VMM, when the software installation is upgraded
to run on hardware with enhancements in the processor’s VMX capabilities. Another
advantage could be that a single software installation image, consisting of a VMM and
guests, could be deployed to multiple hardware platforms with varying VMX capabilities. In such cases, the VMM could fall back to a common subset of VMX features
supported by all VMX revisions, or choose to understand the asymmetry of the VMX
capabilities and assign VMs accordingly.
This section outlines some of the considerations to keep in mind when developing an
MP-aware VMM.
27.8.1
Initialization
Before enabling VMX, an MP-aware VMM must check to make sure that all processors
in the system are compatible and support features required. This can be done by:
•
Checking the CPUID on each logical processor to ensure VMX is supported and
that the overall feature set of each logical processor is compatible.
•
•
Checking VMCS revision identifiers on each logical processor.
Checking each of the “allowed-1” or “allowed-0” fields of the VMX capability
MSR’s on each processor.
27.8.2
Moving a VMCS Between Processors
An MP-aware VMM is free to assign any logical processor to a VM. But for performance considerations, moving a guest VMCS to another logical processor is slower
than resuming that guest VMCS on the same logical processor. Certain VMX performance features (such as caching of portions of the VMCS in the processor) are optimized for a guest VMCS that runs on the same logical processor.
The reasons are:
•
To restart a guest on the same logical processor, a VMM can use VMRESUME.
VMRESUME is expected to be faster than VMLAUNCH in general.
•
To migrate a VMCS to another logical processor, a VMM must use the sequence of
VMCLEAR, VMPTRLD and VMLAUNCH.
•
Operations involving VMCLEAR can impact performance negatively. See
Section 21.10.3.
27-16 Vol. 3B
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
A VMM scheduler should make an effort to schedule a guest VMCS to run on the
logical processor where it last ran. Such a scheduler might also benefit from doing
lazy VMCLEARs (that is: performing a VMCLEAR on a VMCS only when the scheduler
knows the VMCS is being moved to a new logical processor). The remainder of this
section describes the steps a VMM must take to move a VMCS from one processor to
another.
A VMM must check the VMCS revision identifier in the VMX capability MSR
IA32_VMX_BASIC to determine if the VMCS regions are identical between all logical
processors. If the VMCS regions are identical (same revision ID) the following
sequence can be used to move or copy the VMCS from one logical processor to
another:
•
Perform a VMCLEAR operation on the source logical processor. This ensures that
all VMCS data that may be cached by the processor are flushed to memory.
•
Copy the VMCS region from one memory location to another location. This is an
optional step assuming the VMM wishes to relocate the VMCS or move the VMCS
to another system.
•
Perform a VMPTRLD of the physical address of VMCS region on the destination
processor to establish its current VMCS pointer.
If the revision identifiers are different, each field must be copied to an intermediate
structure using individual reads (VMREAD) from the source fields and writes
(VMWRITE) to destination fields. Care must be taken on fields that are hard-wired to
certain values on some processor implementations.
27.8.3
Paired Index-Data Registers
A VMM may need to virtualize hardware that is visible to software using paired indexdata registers. Paired index-data register interfaces, such as those used in PCI (CF8,
CFC), require special treatment in cases where a VM performing writes to these pairs
can be moved during execution. In this case, the index (e.g. CF8) should be part of
the virtualized state. If the VM is moved during execution, writes to the index should
be redone so subsequent data reads/writes go to the right location.
27.8.4
External Data Structures
Certain fields in the VMCS point to external data structures (for example: the MSR
bitmap, the I/O bitmaps). If a logical processor is in VMX non-root operation, none of
the external structures referenced by that logical processor's current VMCS should be
modified by any logical processor or DMA. Before updating one of these structures,
the VMM must ensure that no logical processor whose current VMCS references the
structure is in VMX non-root operation.
If a VMM uses multiple VMCS with each VMCS using separate external structures,
and these structures must be kept synchronized, the VMM must apply the same care
to updating these structures.
Vol. 3B 27-17
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
27.8.5
CPUID Emulation
CPUID reports information that is used by OS and applications to detect hardware
features. It also provides multi-threading/multi-core configuration information. For
example, MP-aware OSs rely on data reported by CPUID to discover the topology of
logical processors in a platform (see Section 8.9, “Programming Considerations for
Hardware Multi-Threading Capable Processors,” in the Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 3A).
If a VMM is to support asymmetric allocation of logical processor resources to guest
OSs that are MP aware, then the VMM must emulate CPUID for its guests. The emulation of CPUID by the VMM must ensure the guest’s view of CPUID leaves are consistent with the logical processor allocation committed by the VMM to each guest OS.
27.9
32-BIT AND 64-BIT GUEST ENVIRONMENTS
For the most part, extensions provided by VMX to support virtualization are orthogonal to the extensions provided by Intel 64 architecture. There are considerations
that impact VMM designs. These are described in the following subsections.
27.9.1
Operating Modes of Guest Environments
For Intel 64 processors, VMX operation supports host and guest environments that
run in IA-32e mode or without IA-32e mode. VMX operation also supports host and
guest environments on IA-32 processors.
A VMM entering VMX operation while IA-32e mode is active is considered to be an
IA-32e mode host. A VMM entering VMX operation while IA-32e mode is not activated
or not available is referred to as a 32-bit VMM. The type of guest operations such
VMMs support are summarized in Table 27-1.
Table 27-1. Operating Modes for Host and Guest Environments
Capability
Guest Operation
in IA-32e mode
Guest Operation
Not Requiring IA-32e Mode
IA-32e mode VMM
Yes
Yes
32-bit VMM
Not supported
Yes
A VM exit may occur to an IA-32e mode guest in either 64-bit sub-mode or compatibility sub-mode of IA-32e mode. VMMs may resume guests in either mode. The submode in which an IA-32e mode guest resumes VMX non-root operation is determined
by the attributes of the code segment which experienced the VM exit. If CS.L = 1,
the guest is executing in 64-bit mode; if CS.L = 0, the guest is executing in compatibility mode (see Section 27.9.5).
27-18 Vol. 3B
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
Not all of an IA-32e mode VMM must run in 64-bit mode. While some parts of an
IA-32e mode VMM must run in 64-bit mode, there are only a few restrictions
preventing a VMM from executing in compatibility mode. The most notable restriction
is that most VMX instructions cause exceptions when executed in compatibility mode.
27.9.2
Handling Widths of VMCS Fields
Individual VMCS control fields must be accessed using VMREAD or VMWRITE instructions. Outside of 64-Bit mode, VMREAD and VMWRITE operate on 32 bits of data. The
widths of VMCS control fields may vary depending on whether a processor supports
Intel 64 architecture.
Many VMCS fields are architected to extend transparently on processors supporting
Intel 64 architecture (64 bits on processors that support Intel 64 architecture, 32 bits
on processors that do not). Some VMCS fields are 64-bits wide regardless of whether
the processor supports Intel 64 architecture or is in IA-32e mode.
27.9.2.1
Natural-Width VMCS Fields
Many VMCS fields operate using natural width. Such fields return (on reads) and set
(on writes) 32-bits when operating in 32-bit mode and 64-bits when operating in
64-bit mode. For the most part, these fields return the naturally expected data
widths. The “Guest RIP” field in the VMCS guest-state area is an example of this type
of field.
27.9.2.2
64-Bit VMCS Fields
Unlike natural width fields, these fields are fixed to 64-bit width on all processors.
When in 64-bit mode, reads of these fields return 64-bit wide data and writes to
these fields write 64-bits. When outside of 64-bit mode, reads of these fields return
the low 32-bits and writes to these fields write the low 32-bits and zero the upper
32-bits. Should a non-IA-32e mode host require access to the upper 32-bits of these
fields, a separate VMCS encoding is used when issuing VMREAD/VMWRITE instructions.
The VMCS control field “MSR bitmap address” (which contains the physical address of
a region of memory which specifies which MSR accesses should generate VM-exits) is
an example of this type of field. Specifying encoding 00002004H to VMREAD returns
the lower 32-bits to non-IA-32e mode hosts and returns 64-bits to 64-bit hosts. The
separate encoding 00002005H returns only the upper 32-bits.
27.9.3
IA-32e Mode Hosts
An IA-32e mode host is required to support 64-bit guest environments. Because activating IA-32e mode currently requires that paging be disabled temporarily and VMX
entry requires paging to be enabled, IA-32e mode must be enabled before entering
Vol. 3B 27-19
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
VMX operation. For this reason, it is not possible to toggle in and out of IA-32e mode
in a VMM.
Section 27.5 describes the steps required to launch a VMM. An IA-32e mode host is
also required to set the “host address-space size” VMCS VM-exit control to 1. The
value of this control is then loaded in the IA32_EFER.LME/LMA and CS.L bits on each
VM exit. This establishes a 64-bit host environment as execution transfers to the
VMM entry point. At a minimum, the entry point is required to be in a 64-bit code
segment. Subsequently, the VMM can, if it chooses, switch to 32-bit compatibility
mode on a code-segment basis (see Section 27.9.1). Note, however, that VMX
instructions other than VMCALL are not supported in compatibility mode; they
generate an invalid opcode exception if used.
The following VMCS controls determine the value of IA32_EFER when a VM exit
occurs: the “host address-space size” control (described above), the “load
IA32_EFER” VM-exit control, the “VM-exit MSR-load count,” and the “VM-exit MSRload address” (see Section 24.3).
If the “load IA32_EFER” VM-exit control is 1, the value of the LME and LMA bits in the
IA32_EFER field in the host-state area must be the value of the “host address-space
size” VM-exit control.
The loading of IA32_EFER.LME/LMA and CS.L bits established by the “host addressspace size” control precede any loading of the IA32_EFER MSR due from the VM-exit
MSR-load area. If IA32_EFER is specified in the VM-exit MSR-load area, the value of
the LME bit in the load image of IA32_EFER should match the setting of the “host
address-space size” control. Otherwise the attempt to modify the LME bit (while
paging is enabled) will lead to a VMX-abort. However, IA32_EFER.LMA is always set
by the processor to equal IA32_EFER.LME & CR0.PG; the value specified for LMA in
the load image of the IA32_EFER MSR is ignored. For these and performance
reasons, VMM writers may choose to not use the VM-exit/entry MSR-load/save areas
for IA32_EFER.
On a VMM teardown, VMX operation should be exited before deactivating IA-32e
mode if the latter is required.
27.9.4
IA-32e Mode Guests
A 32-bit guest can be launched by either IA-32e-mode hosts or non-IA-32e-mode
hosts. A 64-bit guests can only be launched by a IA-32e-mode host.
In addition to the steps outlined in Section 27.6, VMM writers need to:
•
Set the “IA-32e-mode guest” VM-entry control to 1 in the VMCS to assure
VM-entry (VMLAUNCH or VMRESUME) will establish a 64-bit (or 32-bit
compatible) guest operating environment.
•
Enable paging (CR0.PG) and PAE mode (CR4.PAE) to assure VM-entry to a 64-bit
guest will succeed.
27-20 Vol. 3B
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
•
Ensure that the host to be in IA-32e mode (the IA32_EFER.LMA must be set to 1)
and the setting of the VM-exit “host address-space size” control bit in the VMCS
must also be set to 1.
If each of the above conditions holds true, then VM-entry will copy the value of the
VM-entry “IA-32e-mode guest” control bit into the guests IA32_EFER.LME bit, which
will result in subsequent activation of IA-32e mode. If any of the above conditions is
false, the VM-entry will fail and load state from the host-state area of the working
VMCS as if a VM exit had occurred (see Section 23.7).
The following VMCS controls determine the value of IA32_EFER on a VM entry: the
“IA-32e-mode guest” VM-entry control (described above), the “load IA32_EFER” VMentry control, the “VM-entry MSR-load count,” and the “VM-entry MSR-load address”
(see Section 23.4).
If the “load IA32_EFER” VM-entry control is 1, the value of the LME and LMA bits in
the IA32_EFER field in the guest-state area must be the value of the “IA-32e-mode
guest” VM-exit control. Otherwise, the VM entry fails.
The loading of IA32_EFER.LME bit (described above) precedes any loading of the
IA32_EFER MSR from the VM-entry MSR-load area of the VMCS. If loading of
IA32_EFER is specified in the VM-entry MSR-load area, the value of the LME bit in the
load image should be match the setting of the “IA-32e-mode guest” VM-entry
control. Otherwise, the attempt to modify the LME bit (while paging is enabled)
results in a failed VM entry. However, IA32_EFER.LMA is always set by the processor
to equal IA32_EFER.LME & CR0.PG; the value specified for LMA in the load image of
the IA32_EFER MSR is ignored. For these and performance reasons, VMM writers
may choose to not use the VM-exit/entry MSR-load/save areas for IA32_EFER MSR.
Note that the VMM can control the processor’s architectural state when transferring
control to a VM. VMM writers may choose to launch guests in protected mode and
subsequently allow the guest to activate IA-32e mode or they may allow guests to
toggle in and out of IA-32e mode. In this case, the VMM should require VM exit on
accesses to the IA32_EFER MSR to detect changes in the operating mode and modify
the VM-entry “IA-32e-mode guest” control accordingly.
A VMM should save/restore the extended (full 64-bit) contents of the guest generalpurpose registers, the new general-purpose registers (R8-R15) and the SIMD registers introduced in 64-bit mode should it need to modify these upon VM exit.
27.9.5
32-Bit Guests
To launch or resume a 32-bit guest, VMM writers can follow the steps outlined in
Section 27.6, making sure that the “IA-32e-mode guest” VM-entry control bit is set
to 0. Then the “IA-32e-mode guest” control bit is copied into the guest
IA32_EFER.LME bit, establishing IA32_EFER.LMA as 0.
Vol. 3B 27-21
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
27.10
HANDLING MODEL SPECIFIC REGISTERS
Model specific registers (MSR) provide a wide range of functionality. They affect
processor features, control the programming interfaces, or are used in conjunction
with specific instructions. As part of processor virtualization, a VMM may wish to
protect some or all MSR resources from direct guest access.
VMX operation provides the following features to virtualize processor MSRs.
27.10.1
Using VM-Execution Controls
Processor-based VM-execution controls provide two levels of support for handling
guest access to processor MSRs using RDMSR and WRMSR:
•
MSR bitmaps: In VMX implementations that support a 1-setting (see Appendix
G) of the user-MSR-bitmaps execution control bit, MSR bitmaps can be used to
provide flexibility in managing guest MSR accesses. The MSR-bitmap-address in
the guest VMCS can be programmed by VMM to point to a bitmap region which
specifies VM-exit behavior when reading and writing individual MSRs.
MSR bitmaps form a 4-KByte region in physical memory and are required to be
aligned to a 4-KByte boundary. The first 1-KByte region manages read control of
MSRs in the range 00000000H-00001FFFH; the second 1-KByte region covers
read control of MSR addresses in the range C0000000H-C0001FFFH. The bitmaps
for write control of these MSRs are located in the 2-KByte region immediately
following the read control bitmaps. While the MSR bitmap address is part of
VMCS, the MSR bitmaps themselves are not. This implies MSR bitmaps are not
accessible through VMREAD and VMWRITE instructions but rather by using
ordinary memory writes. Also, they are not specially cached by the processor and
may be placed in normal cache-coherent memory by the VMM.
When MSR bitmap addresses are properly programmed and the use-MSR-bitmap
control (see Section 21.6.2) is set, the processor consults the associated bit in
the appropriate bitmap on guest MSR accesses to the corresponding MSR and
causes a VM exit if the bit in the bitmap is set. Otherwise, the access is permitted
to proceed. This level of protection may be utilized by VMMs to selectively allow
guest access to some MSRs while virtualizing others.
•
Default MSR protection: If the use-MSR-bitmap control is not set, an attempt
by a guest to access any MSR causes a VM exit. This also occurs for any attempt
to access an MSR outside the ranges identified above (even if the use-MSRbitmap control is set).
VM exits due to guest MSR accesses may be identified by the VMM through VM-exit
reason codes. The MSR-read exit reason implies guest software attempted to read an
MSR protected either by default or through MSR bitmaps. The MSR-write exit reason
implies guest software attempting to write a MSR protected through the VM-execution controls. Upon VM exits caused by MSR accesses, the VMM may virtualize the
guest MSR access through emulation of RDMSR/WRMSR.
27-22 Vol. 3B
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
27.10.2
Using VM-Exit Controls for MSRs
If a VMM allows its guest to access MSRs directly, the VMM may need to store guest
MSR values and load host MSR values for these MSRs on VM exits. This is especially
true if the VMM uses the same MSRs while in VMX root operation.
A VMM can use the VM-exit MSR-store-address and the VM-exit MSR-store-count exit
control fields (see Section 21.7.2) to manage how MSRs are stored on VM exits. The
VM-exit MSR-store-address field contains the physical address (16-byte aligned) of
the VM-exit MSR-store area (a table of entries with 16 bytes per entry). Each table
entry specifies an MSR whose value needs to be stored on VM exits. The VM-exit
MSR-store-count contains the number of entries in the table.
Similarly the VM-exit MSR-load-address and VM-exit MSR-load-count fields point to
the location and size of the VM-exit MSR load area. The entries in the VM-exit MSRload area contain the host expected values of specific MSRs when a VM exit occurs.
Upon VM-exit, bits 127:64 of each entry in the VM-exit MSR-store area is updated
with the contents of the MSR indexed by bits 31:0. Also, bits 127:64 of each entry in
the VM-exit MSR-load area is updated by loading with values from bits 127:64 the
contents of the MSR indexed by bits 31:0.
27.10.3
Using VM-Entry Controls for MSRs
A VMM may require specific MSRs to be loaded explicitly on VM entries while
launching or resuming guest execution. The VM-entry MSR-load-address and
VM-entry MSR-load-count entry control fields determine how MSRs are loaded on
VM-entries. The VM-entry MSR-load-address and count fields are similar in structure
and function to the VM-exit MSR-load address and count fields, except the MSR
loading is done on VM-entries.
27.10.4
Handling Special-Case MSRs and Instructions
A number of instructions make use of designated MSRs in their operation. The VMM
may need to consider saving the states of those MSRs. Instructions that merit such
consideration include SYSENTER/SYSEXIT, SYSCALL/SYSRET, SWAPGS.
27.10.4.1 Handling IA32_EFER MSR
The IA32_EFER MSR includes bit fields that allow system software to enable
processor features. For example: the SCE bit enables SYSCALL/SYSRET and the NXE
bit enables the execute-disable bits in the paging-structure entries.
VMX provides hardware support to load the IA32_EFER MSR on VMX transitions and
to save it on VM exits. Because of this, VMM software need not use the RDMSR and
WRMSR instruction to give the register different values during host and guest execution.
Vol. 3B 27-23
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
27.10.4.2 Handling the SYSENTER and SYSEXIT Instructions
The SYSENTER and SYSEXIT instructions use three dedicated MSRs
(IA32_SYSENTER_CS, IA32_SYSENTER_ESP and IA32_SYSENTER_EIP) to manage
fast system calls. These MSRs may be utilized by both the VMM and the guest OS to
manage system calls in VMX root operation and VMX non-root operation respectively.
VM entries load these MSRs from fields in the guest-state area of the VMCS. VM exits
save the values of these MSRs into those fields and loads the MSRs from fields in the
host-state area.
27.10.4.3 Handling the SYSCALL and SYSRET Instructions
The SYSCALL/SYSRET instructions are similar to SYSENTER/SYSEXIT but are
designed to operate within the context of a 64-bit flat code segment. They are available only in 64-bit mode and only when the SCE bit of the IA32_EFER MSR is set.
SYSCALL/SYSRET invocations can occur from either 32-bit compatibility mode application code or from 64-bit application code. Three related MSR registers
(IA32_STAR, IA32_LSTAR, IA32_FMASK) are used in conjunction with fast system
calls/returns that use these instructions.
64-Bit hosts which make use of these instructions in the VMM environment will need
to save the guest state of the above registers on VM exit, load the host state, and
restore the guest state on VM entry. One possible approach is to use the VM-exit
MSR-save and MSR-load areas and the VM-entry MSR-load area defined by controls
in the VMCS. A disadvantage to this approach, however, is that the approach results
in the unconditional saving, loading, and restoring of MSR registers on each VM exit
or VM entry.
Depending on the design of the VMM, it is likely that many VM-exits will require no
fast system call support but the VMM will be burdened with the additional overhead
of saving and restoring MSRs if the VMM chooses to support fast system call
uniformly. Further, even if the host intends to support fast system calls during a
VM-exit, some of the MSR values (such as the setting of the SCE bit in IA32_EFER)
may not require modification as they may already be set to the appropriate value in
the guest.
For performance reasons, a VMM may perform lazy save, load, and restore of these
MSR values on certain VM exits when it is determined that this is acceptable. The
lazy-save-load-restore operation can be carried out “manually” using RDMSR and
WRMSR.
27.10.4.4 Handling the SWAPGS Instruction
The SWAPGS instruction is available only in 64-bit mode. It swaps the contents of
two specific MSRs (IA32_GSBASE and IA32_KERNEL_GSBASE). The IA32_GSBASE
MSR shadows the base address portion of the GS descriptor register; the
IA32_KERNEL_GSBASE MSR holds the base address of the GS segment used by the
kernel (typically it houses kernel structures). SWAPGS is intended for use with fast
27-24 Vol. 3B
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
system calls when in 64-bit mode to allow immediate access to kernel structures on
transition to kernel mode.
Similar to SYSCALL/SYSRET, IA-32e mode hosts which use fast system calls may
need to save, load, and restore these MSR registers on VM exit and VM entry using
the guidelines discussed in previous paragraphs.
27.10.4.5 Implementation Specific Behavior on Writing to Certain MSRs
As noted in Section 23.4 and Section 24.4, a processor may prevent writing to
certain MSRs when loading guest states on VM entries or storing guest states on VM
exits. This is done to ensure consistent operation. The subset and number of MSRs
subject to restrictions are implementation specific. For initial VMX implementations,
there are two MSRs: IA32_BIOS_UPDT_TRIG and IA32_BIOS_SIGN_ID (see
Appendix B).
27.10.5
Handling Accesses to Reserved MSR Addresses
Privileged software (either a VMM or a guest OS) can access a model specific register
by specifying addresses in MSR address space. VMMs, however, must prevent a guest
from accessing reserved MSR addresses in MSR address space.
Consult Appendix B for lists of supported MSRs and their usage. Use the MSR bitmap
control to cause a VM exit when a guest attempts to access a reserved MSR address.
The response to such a VM exit should be to reflect #GP(0) back to the guest.
27.11
HANDLING ACCESSES TO CONTROL REGISTERS
Bit fields in control registers (CR0, CR4) control various aspects of processor operation. The VMM must prevent guests from modifying bits in CR0 or CR4 that are
reserved at the time the VMM is written.
Guest/host masks should be used by the VMM to cause VM exits when a guest
attempts to modify reserved bits. Read shadows should be used to ensure that the
guest always reads the reserved value (usually 0) for such bits. The VMM response to
VM exits due to attempts from a guest to modify reserved bits should be to emulate
the response which the processor would have normally produced (usually a #GP(0)).
27.12
PERFORMANCE CONSIDERATIONS
VMX provides hardware features that may be used for improving processor virtualization performance. VMMs must be designed to use this support properly. The basic
idea behind most of these performance optimizations of the VMM is to reduce the
number of VM exits while executing a guest VM.
Vol. 3B 27-25
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
This section lists ways that VMMs can take advantage of the performance enhancing
features in VMX.
•
Read Access to Control Registers. Analysis of common client workloads with
common PC operating systems in a virtual machine shows a large number of
VM-exits are caused by control register read accesses (particularly CR0). Reads
of CR0 and CR4 does not cause VM exits. Instead, they return values from the
CR0/CR4 read-shadows configured by the VMM in the guest controlling-VMCS
with the guest-expected values.
•
Write Access to Control Registers. Most VMM designs require only certain bits
of the control registers to be protected from direct guest access. Write access to
CR0/CR4 registers can be reduced by defining the host-owned and guest-owned
bits in them through the CR0/CR4 host/guest masks in the VMCS. CR0/CR4 write
values by the guest are qualified with the mask bits. If they change only guestowned bits, they are allowed without causing VM exits. Any write that cause
changes to host-owned bits cause VM exits and need to be handled by the VMM.
•
Access Rights based Page Table protection. For VMM that implement
access-rights-based page table protection, the VMCS provides a CR3 target value
list that can be consulted by the processor to determine if a VM exit is required.
Loading of CR3 with a value matching an entry in the CR3 target-list are allowed
to proceed without VM exits. The VMM can utilize the CR3 target-list to save
page-table hierarchies whose state is previously verified by the VMM.
•
Page-fault handling. Another common cause for a VM exit is due to page-faults
induced by guest address remapping done through virtual memory virtualization.
VMX provides page-fault error-code mask and match fields in the VMCS to filter
VM exits due to page-faults based on their cause (reflected in the error-code).
27.13
USE OF THE VMX-PREEMPTION TIMER
The VMX-preemption timer allows VMM software to preempt guest VM execution
after a specified amount of time. Typical VMX-preemption timer usage is to program
the initial VM quantum into the timer, save the timer value on each successive VMexit (using the VM-exit control “save preemption timer value”) and run the VM until
the timer expires.
In an alternative scenario, the VMM may use another timer (e.g. the TSC) to track
the amount of time the VM has run while still using the VMX-preemption timer for VM
preemption. In this scenario the VMM would not save the VMX-preemption timer on
each VM-exit but instead would reload the VMX-preemption timer with initial VM
quantum less the time the VM has already run. This scenario includes all the VMentry and VM-exit latencies in the VM run time.
In both scenarios, on each successive VM-entry the VMX-preemption timer contains
a smaller value until the VM quantum ends. If the VMX-preemption timer is loaded
with a value smaller than the VM-entry latency then the VM will not execute any
27-26 Vol. 3B
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
instructions before the timer expires. The VMM must ensure the initial VM quantum is
greater than the VM-entry latency; otherwise the VM will make no forward progress.
Vol. 3B 27-27
VIRTUAL-MACHINE MONITOR PROGRAMMING CONSIDERATIONS
27-28 Vol. 3B
CHAPTER 28
VIRTUALIZATION OF SYSTEM RESOURCES
28.1
OVERVIEW
When a VMM is hosting multiple guest environments (VMs), it must monitor potential
interactions between software components using the same system resources. These
interactions can require the virtualization of resources. This chapter describes the
virtualization of system resources. These include: debugging facilities, address
translation, physical memory, and microcode update facilities.
28.2
VIRTUALIZATION SUPPORT FOR DEBUGGING
FACILITIES
The Intel 64 and IA-32 debugging facilities (see Chapter 16) provide breakpoint
instructions, exception conditions, register flags, debug registers, control registers
and storage buffers for functions related to debugging system and application software. In VMX operation, a VMM can support debugging system and application software from within virtual machines if the VMM properly virtualizes debugging
facilities. The following list describes features relevant to virtualizing these facilities.
•
The VMM can program the exception-bitmap (see Section 21.6.3) to ensure it
gets control on debug functions (like breakpoint exceptions occurring while
executing guest code such as INT3 instructions). Normally, debug exceptions
modify debug registers (such as DR6, DR7, IA32_DEBUGCTL). However, if debug
exceptions cause VM exits, exiting occurs before register modification.
•
The VMM may utilize the VM-entry event injection facilities described in Section
23.5 to inject debug or breakpoint exceptions to the guest. See Section 28.2.1
for a more detailed discussion.
•
The MOV-DR exiting control bit in the processor-based VM-execution control field
(see Section 21.6.2) can be enabled by the VMM to cause VM exits on explicit
guest access of various processor debug registers (for example, MOV to/from
DR0-DR7). These exits would always occur on guest access of DR0-DR7 registers
regardless of the values in CPL, DR4.DE or DR7.GD. Since all guest task switches
cause VM exits, a VMM can control any indirect guest access or modification of
debug registers during guest task switches.
•
Guest software access to debug-related model-specific registers (such as
IA32_DEBUGCTL MSR) can be trapped by the VMM through MSR access control
features (such as the MSR-bitmaps that are part of processor-based VMexecution controls). See Section 27.10 for details on MSR virtualization.
Vol. 3B 28-1
VIRTUALIZATION OF SYSTEM RESOURCES
•
Debug registers such as DR7 and the IA32_DEBUGCTL MSR may be explicitly
modified by the guest (through MOV-DR or WRMSR instructions) or modified
implicitly by the processor as part of generating debug exceptions. The current
values of DR7 and the IA32_DEBUGCTL MSR are saved to guest-state area of
VMCS on every VM exit. Pending debug exceptions are debug exceptions that are
recognized by the processor but not yet delivered. See Section 23.6.3 for details
on pending debug exceptions.
•
DR7 and the IA32-DEBUGCTL MSR are loaded from values in the guest-state area
of the VMCS on every VM entry. This allows the VMM to properly virtualize debug
registers when injecting debug exceptions to guest. Similarly, the RFLAGS1
register is loaded on every VM entry (or pushed to stack if injecting a virtual
event) from guest-state area of the VMCS. Pending debug exceptions are also
loaded from guest-state area of VMCS so that they may be delivered after VM
entry is completed.
28.2.1
Debug Exceptions
If a VMM emulates a guest instruction that would encounter a debug trap (single step
or data or I/O breakpoint), it should cause that trap to be delivered. The VMM should
not inject the debug exception using VM-entry event injection, but should set the
appropriate bits in the pending debug exceptions field. This method will give the trap
the right priority with respect to other events. (If the exception bitmap was
programmed to cause VM exits on debug exceptions, the debug trap will cause a VM
exit. At this point, the trap can be injected during VM entry with the proper priority.)
There is a valid pending debug exception if the BS bit (see Table 21-4) is set, regardless of the values of RFLAGS.TF or IA32_DEBUGCTL.BTF. The values of these bits do
not impact the delivery of pending debug exceptions.
VMMs should exercise care when emulating a guest write (attempted using WRMSR)
to IA32_DEBUGCTL to modify BTF if this is occurring with RFLAGS.TF = 1 and after a
MOV SS or POP SS instruction (for example: while debug exceptions are blocked).
Note the following:
•
Normally, if WRMSR clears BTF while RFLAGS.TF = 1 and with debug exceptions
blocked, a single-step trap will occur after WRMSR. A VMM emulating such an
instruction should set the BS bit (see Table 21-4) in the pending debug
exceptions field before VM entry.
•
Normally, if WRMSR sets BTF while RFLAGS.TF = 1 and with debug exceptions
blocked, neither a single-step trap nor a taken-branch trap can occur after
WRMSR. A VMM emulating such an instruction should clear the BS bit (see Table
21-4) in the pending debug exceptions field before VM entry.
1. This chapter uses the notation RAX, RIP, RSP, RFLAGS, etc. for processor registers because most
processors that support VMX operation also support Intel 64 architecture. For processors that do
not support Intel 64 architecture, this notation refers to the 32-bit forms of those registers
(EAX, EIP, ESP, EFLAGS, etc.).
28-2 Vol. 3B
VIRTUALIZATION OF SYSTEM RESOURCES
28.3
MEMORY VIRTUALIZATION
VMMs must control physical memory to ensure VM isolation and to remap guest
physical addresses in host physical address space for virtualization. Memory virtualization allows the VMM to enforce control of physical memory and yet support guest
OSs’ expectation to manage memory address translation.
28.3.1
Processor Operating Modes & Memory Virtualization
Memory virtualization is required to support guest execution in various processor
operating modes. This includes: protected mode with paging, protected mode with
no paging, real-mode and any other transient execution modes. VMX allows guest
operation in protected-mode with paging enabled and in virtual-8086 mode (with
paging enabled) to support guest real-mode execution. Guest execution in transient
operating modes (such as in real mode with one or more segment limits greater than
64-KByte) must be emulated by the VMM.
Since VMX operation requires processor execution in protected mode with paging
(through CR0 and CR4 fixed bits), the VMM may utilize paging structures to support
memory virtualization. To support guest real-mode execution, the VMM may establish a simple flat page table for guest linear to host physical address mapping.
Memory virtualization algorithms may also need to capture other guest operating
conditions (such as guest performing A20M# address masking) to map the resulting
20-bit effective guest physical addresses.
28.3.2
Guest & Host Physical Address Spaces
Memory virtualization provides guest software with contiguous guest physical
address space starting zero and extending to the maximum address supported by
the guest virtual processor’s physical address width. The VMM utilizes guest physical
to host physical address mapping to locate all or portions of the guest physical
address space in host memory. The VMM is responsible for the policies and algorithms for this mapping which may take into account the host system physical
memory map and the virtualized physical memory map exposed to a guest by the
VMM. The memory virtualization algorithm needs to accommodate various guest
memory uses (such as: accessing DRAM, accessing memory-mapped registers of
virtual devices or core logic functions and so forth). For example:
•
To support guest DRAM access, the VMM needs to map DRAM-backed guest
physical addresses to host-DRAM regions. The VMM also requires the guest to
host memory mapping to be at page granularity.
•
Virtual devices (I/O devices or platform core logic) emulated by the VMM may
claim specific regions in the guest physical address space to locate memorymapped registers. Guest access to these virtual registers may be configured to
cause page-fault induced VM-exits by marking these regions as always not
Vol. 3B 28-3
VIRTUALIZATION OF SYSTEM RESOURCES
present. The VMM may handle these VM exits by invoking appropriate virtual
device emulation code.
28.3.3
Virtualizing Virtual Memory by Brute Force
VMX provides the hardware features required to fully virtualize guest virtual memory
accesses. VMX allows the VMM to trap guest accesses to the PAT (Page Attribute
Table) MSR and the MTRR (Memory Type Range Registers). This control allows the
VMM to virtualize the specific memory type of a guest memory. The VMM may control
caching by controlling the guest CR0.CRD and CR0.NW bits, as well as by trapping
guest execution of the INVD instruction. The VMM can trap guest CR3 loads and
stores, and it may trap guest execution of INVLPG.
Because a VMM must retain control of physical memory, it must also retain control
over the processor’s address-translation mechanisms. Specifically, this means that
only the VMM can access CR3 (which contains the base of the page directory) and can
execute INVLPG (the only other instruction that directly manipulates the TLB).
At the same time that the VMM controls address translation, a guest operating
system will also expect to perform normal memory management functions. It will
access CR3, execute INVLPG, and modify (what it believes to be) page directories
and page tables. Virtualization of address translation must tolerate and support
guest attempts to control address translation.
A simple-minded way to do this would be to ensure that all guest attempts to access
address-translation hardware trap to the VMM where such operations can be properly
emulated. It must ensure that accesses to page directories and page tables also get
trapped. This may be done by protecting these in-memory structures with conventional page-based protection. The VMM can do this because it can locate the page
directory because its base address is in CR3 and the VMM receives control on any
change to CR3; it can locate the page tables because their base addresses are in the
page directory.
Such a straightforward approach is not necessarily desirable. Protection of the inmemory translation structures may be cumbersome. The VMM may maintain these
structures with different values (e.g., different page base addresses) than guest software. This means that there must be traps on guest attempt to read these structures
and that the VMM must maintain, in auxiliary data structures, the values to return to
these reads. There must also be traps on modifications to these structures even if the
translations they effect are never used. All this implies considerable overhead that
should be avoided.
28.3.4
Alternate Approach to Memory Virtualization
Guest software is allowed to freely modify the guest page-table hierarchy without
causing traps to the VMM. Because of this, the active page-table hierarchy might not
always be consistent with the guest hierarchy. Any potential problems arising from
28-4 Vol. 3B
VIRTUALIZATION OF SYSTEM RESOURCES
inconsistencies can be solved using techniques analogous to those used by the
processor and its TLB.
This section describes an alternative approach that allows guest software to freely
access page directories and page tables. Traps occur on CR3 accesses and executions
of INVLPG. They also occur when necessary to ensure that guest modifications to the
translation structures actually take effect. The software mechanisms to support this
approach are collectively called virtual TLB. This is because they emulate the functionality of the processor’s physical translation look-aside buffer (TLB).
The basic idea behind the virtual TLB is similar to that behind the processor TLB.
While the page-table hierarchy defines the relationship between physical to linear
address, it does not directly control the address translation of each memory access.
Instead, translation is controlled by the TLB, which is occasionally filled by the
processor with translations derived from the page-table hierarchy. With a virtual TLB,
the page-table hierarchy established by guest software (specifically, the guest operating system) does not control translation, either directly or indirectly. Instead,
translation is controlled by the processor (through its TLB) and by the VMM (through
a page-table hierarchy that it maintains).
Specifically, the VMM maintains an alternative page-table hierarchy that effectively
caches translations derived from the hierarchy maintained by guest software. The
remainder of this document refers to the former as the active page-table hierarchy
(because it is referenced by CR3 and may be used by the processor to load its TLB)
and the latter as the guest page-table hierarchy (because it is maintained by guest
software). The entries in the active hierarchy may resemble the corresponding
entries in the guest hierarchy in some ways and may differ in others.
Guest software is allowed to freely modify the guest page-table hierarchy without
causing VM exits to the VMM. Because of this, the active page-table hierarchy might
not always be consistent with the guest hierarchy. Any potential problems arising
from any inconsistencies can be solved using techniques analogous to those used by
the processor and its TLB. Note the following:
•
Suppose the guest page-table hierarchy allows more access than active hierarchy
(for example: there is a translation for a linear address in the guest hierarchy but
not in the active hierarchy); this is analogous to a situation in which the TLB
allows less access than the page-table hierarchy. If an access occurs that would
be allowed by the guest hierarchy but not the active one, a page fault occurs; this
is analogous to a TLB miss. The VMM gains control (as it handles all page faults)
and can update the active page-table hierarchy appropriately; this corresponds
to a TLB fill.
•
Suppose the guest page-table hierarchy allows less access than the active
hierarchy; this is analogous to a situation in which the TLB allows more access
than the page-table hierarchy. This situation can occur only if the guest operating
system has modified a page-table entry to reduce access (for example: by
marking it not-present). Because the older, more permissive translation may
have been cached in the TLB, the processor is architecturally permitted to use the
older translation and allow more access. Thus, the VMM may (through the active
page-table hierarchy) also allow greater access. For the new, less permissive
Vol. 3B 28-5
VIRTUALIZATION OF SYSTEM RESOURCES
translation to take effect, guest software should flush any older translations from
the TLB either by executing INVLPG or by loading CR3. Because both these
operations will cause a trap to the VMM, the VMM will gain control and can
remove from the active page-table hierarchy the translations indicated by guest
software (the translation of a specific linear address for INVLPG or all translations
for a load of CR3).
As noted previously, the processor reads the page-table hierarchy to cache translations in the TLB. It also writes to the hierarchy to main the accessed (A) and dirty (D)
bits in the PDEs and PTEs. The virtual TLB emulates this behavior as follows:
•
When a page is accessed by guest software, the A bit in the corresponding PTE
(or PDE for a 4-MByte page) in the active page-table hierarchy will be set by the
processor (the same is true for PDEs when active page tables are accessed by the
processor). For guest software to operate properly, the VMM should update the A
bit in the guest entry at this time. It can do this reliably if it keeps the active PTE
(or PDE) marked not-present until it has set the A bit in the guest entry.
•
When a page is written by guest software, the D bit in the corresponding PTE (or
PDE for a 4-MByte page) in the active page-table hierarchy will be set by the
processor. For guest software to operate properly, the VMM should update the D
bit in the guest entry at this time. It can do this reliably if it keeps the active PTE
(or PDE) marked read-only until it has set the D bit in the guest entry. This
solution is valid for guest software running at privilege level 3; support for more
privileged guest software is described in Section 28.3.5.
28.3.5
Details of Virtual TLB Operation
This section describes in more detail how a VMM could support a virtual TLB. It
explains how an active page-table hierarchy is initialized and how it is maintained in
response to page faults, uses of INVLPG, and accesses to CR3. The mechanisms
described here are the minimum necessary. They may not result in the best performance.
28-6 Vol. 3B
VIRTUALIZATION OF SYSTEM RESOURCES
"Virtual TLB"
Active Page-Table Hierarchy
Guest Page-Table Hierarchy
Guest
Active
F
CR3
TLB
refill on
TLB miss
set dirty
accessed
PT
F
PD
F
F
CR3
refill on
page fault
set accessed
and dirty bits
PT
PT
INVLPG
MOV to
CR3
task switch
F
PD
F
PT
F
F
INVLPG
MOV to CR3
task switch
PD = page directory
PT = page table
F = page frame
OM19040
Figure 28-1. Virtual TLB Scheme
As noted above, the VMM maintains an active page-table hierarchy for each virtual
machine that it supports. It also maintains, for each machine, values that the
machine expects for control registers CR0, CR2, CR3, and CR4 (they control address
translation). These values are called the guest control registers.
In general, the VMM selects the physical-address space that is allocated to guest
software. The term guest address refers to an address installed by guest software in
the guest CR3, in a guest PDE (as a page table base address or a page base address),
or in a guest PTE (as a page base address). While guest software considers these to
be specific physical addresses, the VMM may map them differently.
28.3.5.1
Initialization of Virtual TLB
To enable the Virtual TLB scheme, the VMCS must be set up to trigger VM exits on:
•
All writes to CR3 (the CR3-target count should be 0) or the paging-mode bits in
CR0 and CR4 (using the CR0 and CR4 guest/host masks)
•
•
Page-fault (#PF) exceptions
Execution of INVLPG
Vol. 3B 28-7
VIRTUALIZATION OF SYSTEM RESOURCES
When guest software first enables paging, the VMM creates an aligned 4-KByte active
page directory that is invalid (all entries marked not-present). This invalid directory
is analogous to an empty TLB.
28.3.5.2
Response to Page Faults
Page faults can occur for a variety of reasons. In some cases, the page fault alerts the
VMM to an inconsistency between the active and guest page-table hierarchy. In such
cases, the VMM can update the former and re-execute the faulting instruction. In
other cases, the hierarchies are already consistent and the fault should be handled
by the guest operating system. The VMM can detect this and use an established
mechanism for raising a page fault to guest software.
The VMM can handle a page fault by following these steps (The steps below assume
the guest is operating in a paging mode without PAE. Analogous steps to handle
address translation using PAE or four-level paging mechanisms can be derived by
VMM developers according to the paging behavior defined in Chapter 3 of the Intel®
64 and IA-32 Architectures Software Developer’s Manual, Volume 3A):
1. First consult the active PDE, which can be located using the upper 10 bits of the
faulting address and the current value of CR3. The active PDE is the source of the
fault if it is marked not present or if its R/W bit and U/S bits are inconsistent with
the attempted guest access (the guest privilege level and the values of CR0.WP
and CR4.SMEP should also be taken into account).
2. If the active PDE is the source of the fault, consult the corresponding guest PDE
using the same 10 bits from the faulting address and the physical address that
corresponds to the guest address in the guest CR3. If the guest PDE would cause
a page fault (for example: it is marked not present), then raise a page fault to the
guest operating system.
The following steps assume that the guest PDE would not have caused a page
fault.
3. If the active PDE is the source of the fault and the guest PDE contains, as pagetable base address (if PS = 0) or page base address (PS = 1), a guest address
that the VMM has chosen not to support; then raise a machine check (or some
other abort) to the guest operating system.
The following steps assume that the guest address in the guest PDE is supported
for the virtual machine.
4. If the active PDE is marked not-present, then set the active PDE to correspond to
guest PDE as follows:
a. If the active PDE contains a page-table base address (if PS = 0), then
allocate an aligned 4-KByte active page table marked completely invalid and
set the page-table base address in the active PDE to be the physical address
of the newly allocated page table.
28-8 Vol. 3B
VIRTUALIZATION OF SYSTEM RESOURCES
b. If the active PDE contains a page base address (if PS = 1), then set the page
base address in the active PDE to be the physical page base address that
corresponds to the guest address in the guest PDE.
c.
Set the P, U/S, and PS bits in the active PDE to be identical to those in the
guest PDE.
d. Set the PWT, PCD, and G bits according to the policy of the VMM.
e. Set A = 1 in the guest PDE.
f.
If D = 1 in the guest PDE or PS = 0 (meaning that this PDE refers to a page
table), then set the R/W bit in the active PDE as in the guest PDE.
g. If D = 0 in the guest PDE, PS = 1 (this is a 4-MByte page), and the attempted
access is a write; then set R/W in the active PDE as in the guest PDE and set
D = 1 in the guest PDE.
h. If D = 0 in the guest PDE, PS = 1, and the attempted access is not a write;
then set R/W = 0 in the active PDE.
i.
After modifying the active PDE, re-execute the faulting instruction.
The remaining steps assume that the active PDE is already marked present.
5. If the active PDE is the source of the fault, the active PDE refers to a 4-MByte
page (PS = 1), the attempted access is a write; D = 0 in the guest PDE, and the
active PDE has caused a fault solely because it has R/W = 0; then set R/W in the
active PDE as in the guest PDE; set D = 1 in the guest PDE, and re-execute the
faulting instruction.
6. If the active PDE is the source of the fault and none of the above cases apply,
then raise a page fault of the guest operating system.
The remaining steps assume that the source of the original page fault is not the
active PDE.
NOTE
It is possible that the active PDE might be causing a fault even
though the guest PDE would not. However, this can happen only if the
guest operating system increased access in the guest PDE and did
not take action to ensure that older translations were flushed from
the TLB. Such translations might have caused a page fault if the
guest software were running on bare hardware.
7. If the active PDE refers to a 4-MByte page (PS = 1) but is not the source of the
fault, then the fault resulted from an inconsistency between the active page-table
hierarchy and the processor’s TLB. Since the transition to the VMM caused an
address-space change and flushed the processor’s TLB, the VMM can simply reexecute the faulting instruction.
The remaining steps assume that PS = 0 in the active and guest PDEs.
Vol. 3B 28-9
VIRTUALIZATION OF SYSTEM RESOURCES
8. Consult the active PTE, which can be located using the next 10 bits of the faulting
address (bits 21–12) and the physical page-table base address in the active PDE.
The active PTE is the source of the fault if it is marked not-present or if its R/W bit
and U/S bits are inconsistent with the attempted guest access (the guest
privilege level and the values of CR0.WP and CR4.SMEP should also be taken into
account).
9. If the active PTE is not the source of the fault, then the fault has resulted from an
inconsistency between the active page-table hierarchy and the processor’s TLB.
Since the transition to the VMM caused an address-space change and flushed the
processor’s TLB, the VMM simply re-executes the faulting instruction.
The remaining steps assume that the active PTE is the source of the fault.
10. Consult the corresponding guest PTE using the same 10 bits from the faulting
address and the physical address that correspond to the guest page-table base
address in the guest PDE. If the guest PTE would cause a page fault (it is marked
not-present), the raise a page fault to the guest operating system.
The following steps assume that the guest PTE would not have caused a page
fault.
11. If the guest PTE contains, as page base address, a physical address that is not
valid for the virtual machine being supported; then raise a machine check (or
some other abort) to the guest operating system.
The following steps assume that the address in the guest PTE is valid for the
virtual machine.
12. If the active PTE is marked not-present, then set the active PTE to correspond to
guest PTE:
a. Set the page base address in the active PTE to be the physical address that
corresponds to the guest page base address in the guest PTE.
b. Set the P, U/S, and PS bits in the active PTE to be identical to those in the
guest PTE.
c.
Set the PWT, PCD, and G bits according to the policy of the VMM.
d. Set A = 1 in the guest PTE.
e. If D = 1 in the guest PTE, then set the R/W bit in the active PTE as in the
guest PTE.
f.
If D = 0 in the guest PTE and the attempted access is a write, then set R/W in
the active PTE as in the guest PTE and set D = 1 in the guest PTE.
g. If D = 0 in the guest PTE and the attempted access is not a write, then set
R/W = 0 in the active PTE.
h. After modifying the active PTE, re-execute the faulting instruction.
The remaining steps assume that the active PTE is already marked present.
13. If the attempted access is a write, D = 0 (not dirty) in the guest PTE and the
active PTE has caused a fault solely because it has R/W = 0 (read-only); then set
28-10 Vol. 3B
VIRTUALIZATION OF SYSTEM RESOURCES
R/W in the active PTE as in the guest PTE, set D = 1 in the guest PTE and reexecute the faulting instruction.
14. If none of the above cases apply, then raise a page fault of the guest operating
system.
28.3.5.3
Response to Uses of INVLPG
Operating-systems can use INVLPG to flush entries from the TLB. This instruction
takes a linear address as an operand and software expects any cached translations
for the address to be flushed. A VMM should set the processor-based VM-execution
control “INVLPG exiting” to 1 so that any attempts by a privileged guest to execute
INVLPG will trap to the VMM. The VMM can then modify the active page-table hierarchy to emulate the desired effect of the INVLPG.
The following steps are performed. Note that these steps are performed only if the
guest invocation of INVLPG would not fault and only if the guest software is running
at privilege level 0:
1. Locate the relevant active PDE using the upper 10 bits of the operand address
and the current value of CR3. If the PDE refers to a 4-MByte page (PS = 1), then
set P = 0 in the PDE.
2. If the PDE is marked present and refers to a page table (PS = 0), locate the
relevant active PTE using the next 10 bits of the operand address (bits 21–12)
and the page-table base address in the PDE. Set P = 0 in the PTE. Examine all
PTEs in the page table; if they are now all marked not-present, de-allocate the
page table and set P = 0 in the PDE (this step may be optional).
28.3.5.4
Response to CR3 Writes
A guest operating system may attempt to write to CR3. Any write to CR3 implies a
TLB flush and a possible page table change. The following steps are performed:
1. The VMM notes the new CR3 value (used later to walk guest page tables) and
emulates the write.
2. The VMM allocates a new PD page, with all invalid entries.
3. The VMM sets actual processor CR3 register to point to the new PD page.
The VMM may, at this point, speculatively fill in VTLB mappings for performance
reasons.
28.4
MICROCODE UPDATE FACILITY
The microcode code update facility may be invoked at various points during the operation of a platform. Typically, the BIOS invokes the facility on all processors during
the BIOS boot process. This is sufficient to boot the BIOS and operating system. As a
Vol. 3B 28-11
VIRTUALIZATION OF SYSTEM RESOURCES
microcode update more current than the system BIOS may be available, system software should provide another mechanism for invoking the microcode update facility.
The implications of the microcode update mechanism on the design of the VMM are
described in this section.
NOTE
Microcode updates must not be performed during VMX non-root
operation. Updates performed in VMX non-root operation may result
in unpredictable system behavior.
28.4.1
Early Load of Microcode Updates
The microcode update facility may be invoked early in the VMM or guest OS boot
process. Loading the microcode update early provides the opportunity to correct
errata affecting the boot process but the technique generally requires a reboot of the
software.
A microcode update may be loaded from the OS or VMM image loader. Typically, such
image loaders do not run on every logical processor, so this method effects only one
logical processor. Later in the VMM or OS boot process, after bringing all application
processors on-line, the VMM or OS needs to invoke the microcode update facility for
all application processors.
Depending on the order of the VMM and the guest OS boot, the microcode update
facility may be invoked by the VMM or the guest OS. For example, if the guest OS
boots first and then loads the VMM, the guest OS may invoke the microcode update
facility on all the logical processors. If a VMM boots before its guests, then the VMM
may invoke the microcode update facility during its boot process. In both cases, the
VMM or OS should invoke the microcode update facilities soon after performing the
multiprocessor startup.
In the early load scenario, microcode updates may be contained in the VMM or OS
image or, the VMM or OS may manage a separate database or file of microcode
updates. Maintaining a separate microcode update image database has the advantage of reducing the number of required VMM or OS releases as a result of microcode
update releases.
28.4.2
Late Load of Microcode Updates
A microcode update may be loaded during normal system operation. This allows
system software to activate the microcode update at anytime without requiring a
system reboot. This scenario does not allow the microcode update to correct errata
which affect the processor’s boot process but does allow high-availability systems to
activate microcode updates without interrupting the availability of the system. In this
late load scenario, either the VMM or a designated guest may load the microcode
update. If the guest is loading the microcode update, the VMM must make sure that
28-12 Vol. 3B
VIRTUALIZATION OF SYSTEM RESOURCES
the entire guest memory buffer (which contains the microcode update image) will not
cause a page fault when accessed.
If the VMM loads the microcode update, then the VMM must have access to the
current set of microcode updates. These updates could be part of the VMM image or
could be contained in a separate microcode update image database (for example: a
database file on disk or in memory). Again, maintaining a separate microcode update
image database has the advantage of reducing the number of required VMM or OS
releases as a result of microcode update releases.
The VMM may wish to prevent a guest from loading a microcode update or may wish
to support the microcode update requested by a guest using emulation (without
actually loading the microcode update). To prevent microcode update loading, the
VMM may return a microcode update signature value greater than the value of
IA32_BIOS_SIGN_ID MSR. A well behaved guest will not attempt to load an older
microcode update. The VMM may also drop the guest attempts to write to
IA32_BIOS_UPDT_TRIG MSR, preventing the guest from loading any microcode
updates. Later, when the guest queries IA32_BIOS_SIGN_ID MSR, the VMM could
emulate the microcode update signature that the guest expects.
In general, loading a microcode update later will limit guest software’s visibility of
features that may be enhanced by a microcode update.
Vol. 3B 28-13
VIRTUALIZATION OF SYSTEM RESOURCES
28-14 Vol. 3B
CHAPTER 29
HANDLING BOUNDARY CONDITIONS IN A VIRTUAL
MACHINE MONITOR
29.1
OVERVIEW
This chapter describes what a VMM must consider when handling exceptions, interrupts, error conditions, and transitions between activity states.
29.2
INTERRUPT HANDLING IN VMX OPERATION
The following bullets summarize VMX support for handling interrupts:
•
Control of processor exceptions. The VMM can get control on specific guest
exceptions through the exception-bitmap in the guest controlling VMCS. The
exception bitmap is a 32-bit field that allows the VMM to specify processor
behavior on specific exceptions (including traps, faults, and aborts). Setting a
specific bit in the exception bitmap implies VM exits will be generated when the
corresponding exception occurs. Any exceptions that are programmed not to
cause VM exits are delivered directly to the guest through the guest IDT. The
exception bitmap also controls execution of relevant instructions such as BOUND,
INTO and INT3. VM exits on page-faults are treated in such a way the page-fault
error code is qualified through the page-fault-error-code mask and match fields
in the VMCS.
•
Control over triple faults. If a fault occurs while attempting to call a doublefault handler in the guest and that fault is not configured to cause a VM exit in the
exception bitmap, the resulting triple fault causes a VM exit.
•
Control of external interrupts. VMX allows both host and guest control of
external interrupts through the “external-interrupt exiting” VM execution control.
If the control is 0, external-interrupts do not cause VM exits and the interrupt
delivery is masked by the guest programmed RFLAGS.IF value.1 If the control is
1, external-interrupts causes VM exits and are not masked by RFLAGS.IF. The
VMM can identify VM exits due to external interrupts by checking the exit reason
for an “external interrupt” (value = 1).
1. This chapter uses the notation RAX, RIP, RSP, RFLAGS, etc. for processor registers because most
processors that support VMX operation also support Intel 64 architecture. For processors that do
not support Intel 64 architecture, this notation refers to the 32-bit forms of those registers
(EAX, EIP, ESP, EFLAGS, etc.).
Vol. 3B 29-1
HANDLING BOUNDARY CONDITIONS IN A VIRTUAL MACHINE MONITOR
•
Control of other events. There is a pin-based VM-execution control that
controls system behavior (exit or no-exit) for NMI events. Most VMM usages will
need handling of NMI external events in the VMM and hence will specify host
control of these events.
Some processors also support a pin-based VM-execution control called “virtual
NMIs.” When this control is set, NMIs cause VM exits, but the processor tracks
guest readiness for virtual NMIs. This control interacts with the “NMI-window
exiting” VM-execution control (see below).
INIT and SIPI events always cause VM exits.
•
Acknowledge interrupt on exit. The “acknowledge interrupt on exit” VM-exit
control in the controlling VMCS controls processor behavior for external interrupt
acknowledgement. If the control is 1, the processor acknowledges the interrupt
controller to acquire the interrupt vector upon VM exit, and stores the vector in
the VM-exit interruption-information field. If the control is 0, the external
interrupt is not acknowledged during VM exit. Since RFLAGS.IF is automatically
cleared on VM exits due to external interrupts, VMM re-enabling of interrupts
(setting RFLAGS.IF = 1) initiates the external interrupt acknowledgement and
vectoring of the external interrupt through the monitor/host IDT.
•
Event-masking Support. VMX captures the masking conditions of specific
events while in VMX non-root operation through the interruptibility-state field in
the guest-state area of the VMCS.
This feature allows proper virtualization of various interrupt blocking states, such
as: (a) blocking of external interrupts for the instruction following STI; (b)
blocking of interrupts for the instruction following a MOV-SS or POP-SS
instruction; (c) SMI blocking of subsequent SMIs until the next execution of RSM;
and (d) NMI/SMI blocking of NMIs until the next execution of IRET or RSM.
INIT and SIPI events are treated specially. INIT assertions are always blocked in
VMX root operation and while in SMM, and unblocked otherwise. SIPI events are
always blocked in VMX root operation.
The interruptibility state is loaded from the VMCS guest-state area on every
VM entry and saved into the VMCS on every VM exit.
•
Event injection. VMX operation allows injecting interruptions to a guest virtual
machine through the use of VM-entry interrupt-information field in VMCS.
Injectable interruptions include external interrupts, NMI, processor exceptions,
software generated interrupts, and software traps. If the interrupt-information
field indicates a valid interrupt, exception or trap event upon the next VM entry;
the processor will use the information in the field to vector a virtual interruption
through the guest IDT after all guest state and MSRs are loaded. Delivery
through the guest IDT emulates vectoring in non-VMX operation by doing the
normal privilege checks and pushing appropriate entries to the guest stack
(entries may include RFLAGS, EIP and exception error code). A VMM with host
control of NMI and external interrupts can use the event-injection facility to
forward virtual interruptions to various guest virtual machines.
29-2 Vol. 3B
HANDLING BOUNDARY CONDITIONS IN A VIRTUAL MACHINE MONITOR
•
Interrupt-window exiting. When set to 1, the “interrupt-window exiting” VMexecution control (Section 21.6.2) causes VM exits when guest RFLAGS.IF is 1
and no other conditions block external interrupts. A VM exit occurs at the
beginning of any instruction at which RFLAGS.IF = 1 and on which the interruptibility state of the guest would allow delivery of an interrupt. For example: when
the guest executes an STI instruction, RFLAGS = 1, and if at the completion of
next instruction the interruptibility state masking due to STI is removed; a
VM exit occurs if the “interrupt-window exiting” VM-execution control is 1. This
feature allows a VMM to queue a virtual interrupt to the guest when the guest is
not in an interruptible state. The VMM can set the “interrupt-window exiting” VMexecution control for the guest and depend on a VM exit to know when the guest
becomes interruptible (and, therefore, when it can inject a virtual interrupt). The
VMM can detect such VM exits by checking for the basic exit reason “interruptwindow” (value = 7). If this feature is not used, the VMM will need to poll and
check the interruptibility state of the guest to deliver virtual interrupts.
•
NMI-window exiting. If the “virtual NMIs” VM-execution is set, the processor
tracks virtual-NMI blocking. The “NMI-window exiting” VM-execution control
(Section 21.6.2) causes VM exits when there is no virtual-NMI blocking. For
example, after execution of the IRET instruction, a VM exit occurs if the “NMIwindow exiting” VM-execution control is 1. This feature allows a VMM to queue a
virtual NMI to a guest when the guest is not ready to receive NMIs. The VMM can
set the “NMI-window exiting” VM-execution control for the guest and depend on
a VM exit to know when the guest becomes ready for NMIs (and, therefore, when
it can inject a virtual NMI). The VMM can detect such VM exits by checking for the
basic exit reason “NMI window” (value = 8). If this feature is not used, the VMM
will need to poll and check the interruptibility state of the guest to deliver virtual
NMIs.
•
VM-exit information. The VM-exit information fields provide details on VM exits
due to exceptions and interrupts. This information is provided through the exitqualification, VM-exit-interruption-information, instruction-length and interruption-error-code fields. Also, for VM exits that occur in the course of vectoring
through the guest IDT, information about the event that was being vectored
through the guest IDT is provided in the IDT-vectoring-information and IDTvectoring-error-code fields. These information fields allow the VMM to identify
the exception cause and to handle it properly.
29.3
EXTERNAL INTERRUPT VIRTUALIZATION
VMX operation allows both host and guest control of external interrupts. While guest
control of external interrupts might be suitable for partitioned usages (different CPU
cores/threads and I/O devices partitioned to independent virtual machines), most
VMMs built upon VMX are expected to utilize host control of external interrupts. The
rest of this section describes a general host-controlled interrupt virtualization architecture for standard PC platforms through the use of VMX supported features.
Vol. 3B 29-3
HANDLING BOUNDARY CONDITIONS IN A VIRTUAL MACHINE MONITOR
With host control of external interrupts, the VMM (or the host OS in a hosted VMM
model) manages the physical interrupt controllers in the platform and the interrupts
generated through them. The VMM exposes software-emulated virtual interrupt
controller devices (such as PIC and APIC) to each guest virtual machine instance.
29.3.1
Virtualization of Interrupt Vector Space
The Intel 64 and IA-32 architectures use 8-bit vectors of which 244 (20H – FFH) are
available for external interrupts. Vectors are used to select the appropriate entry in
the interrupt descriptor table (IDT). VMX operation allows each guest to control its
own IDT. Host vectors refer to vectors delivered by the platform to the processor
during the interrupt acknowledgement cycle. Guest vectors refer to vectors
programmed by a guest to select an entry in its guest IDT. Depending on the I/O
resource management models supported by the VMM design, the guest vector space
may or may not overlap with the underlying host vector space.
•
Interrupts from virtual devices: Guest vector numbers for virtual interrupts
delivered to guests on behalf of emulated virtual devices have no direct relation
to the host vector numbers of interrupts from physical devices on which they are
emulated. A guest-vector assigned for a virtual device by the guest operating
environment is saved by the VMM and utilized when injecting virtual interrupts on
behalf of the virtual device.
•
Interrupts from assigned physical devices: Hardware support for I/O device
assignment allows physical I/O devices in the host platform to be assigned
(direct-mapped) to VMs. Guest vectors for interrupts from direct-mapped
physical devices take up equivalent space from the host vector space, and
require the VMM to perform host-vector to guest-vector mapping for interrupts.
Figure 29-1 illustrates the functional relationship between host external interrupts
and guest virtual external interrupts. Device A is owned by the host and generates
external interrupts with host vector X. The host IDT is set up such that the interrupt
service routine (ISR) for device driver A is hooked to host vector X as normal. VMM
emulates (over device A) virtual device C in software which generates virtual interrupts to the VM with guest expected vector P. Device B is assigned to a VM and generates external interrupts with host vector Y. The host IDT is programmed to hook the
VMM interrupt service routine (ISR) for assigned devices for vector Y, and the VMM
handler injects virtual interrupt with guest vector Q to the VM. The guest operating
system programs the guest to hook appropriate guest driver’s ISR to vectors P
and Q.
29-4 Vol. 3B
HANDLING BOUNDARY CONDITIONS IN A VIRTUAL MACHINE MONITOR
VM
Device Driver C
Guest IDT
Device Driver B
Guest
Vector P
Guest
Vector Q
Guest IDTR
Virtual Interrupt
Virtual Interrupt
Virtual Device C
Emulation
Host IDT
Monitor Handler
Platform Interrupt
Vector Y
Host
Virtual Machine Monitor (VMM)
Host
Host IDTR
Vector X
Device Driver A
Platform Interrupt
Hardware
Device A
Device B
OM19041
Figure 29-1. Host External Interrupts and Guest Virtual Interrupts
29.3.2
Control of Platform Interrupts
To meet the interrupt virtualization requirements, the VMM needs to take ownership
of the physical interrupts and the various interrupt controllers in the platform. VMM
control of physical interrupts may be enabled through the host-control settings of the
“external-interrupt exiting” VM-execution control. To take ownership of the platform
interrupt controllers, the VMM needs to expose the virtual interrupt controller devices
to the virtual machines and restrict guest access to the platform interrupt controllers.
Intel 64 and IA-32 platforms can support three types of external interrupt control
mechanisms: Programmable Interrupt Controllers (PIC), Advanced Programmable
Vol. 3B 29-5
HANDLING BOUNDARY CONDITIONS IN A VIRTUAL MACHINE MONITOR
Interrupt Controllers (APIC), and Message Signaled Interrupts (MSI). The following
sections provide information on the virtualization of each of these mechanisms.
29.3.2.1
PIC Virtualization
Typical PIC-enabled platform implementations support dual 8259 interrupt controllers cascaded as master and slave controllers. They supporting up to 15 possible
interrupt inputs. The 8259 controllers are programmed through initialization
command words (ICWx) and operation command words (OCWx) accessed through
specific I/O ports. The various interrupt line states are captured in the PIC through
interrupt requests, interrupt service routines and interrupt mask registers.
Guest access to the PIC I/O ports can be restricted by activating I/O bitmaps in the
guest controlling-VMCS (activate-I/O-bitmap bit in VM-execution control field set
to 1) and pointing the I/O-bitmap physical addresses to valid bitmap regions. Bits
corresponding to the PIC I/O ports can be cleared to cause a VM exit on guest access
to these ports.
If the VMM is not supporting direct access to any I/O ports from a guest, it can set the
unconditional-I/O-exiting in the VM-execution control field instead of activating I/O
bitmaps. The exit-reason field in VM-exit information allows identification of VM exits
due to I/O access and can provide an exit-qualification to identify details about the
guest I/O operation that caused the VM exit.
The VMM PIC virtualization needs to emulate the platform PIC functionality including
interrupt priority, mask, request and service states, and specific guest programmed
modes of PIC operation.
29.3.2.2
xAPIC Virtualization
Most modern Intel 64 and IA-32 platforms include support for an APIC. While the
standard PIC is intended for use on uniprocessor systems, APIC can be used in either
uniprocessor or multi-processor systems.
APIC based interrupt control consists of two physical components: the interrupt
acceptance unit (Local APIC) which is integrated with the processor, and the interrupt
delivery unit (I/O APIC) which is part of the I/O subsystem. APIC virtualization
involves protecting the platform’s local and I/O APICs and emulating them for the
guest.
29.3.2.3
Local APIC Virtualization
The local APIC is responsible for the local interrupt sources, interrupt acceptance,
dispensing interrupts to the logical processor, and generating inter-processor interrupts. Software interacts with the local APIC by reading and writing its memorymapped registers residing within a 4-KByte uncached memory region with base
address stored in the IA32_APIC_BASE MSR. Since the local APIC registers are
memory-mapped, the VMM can utilize memory virtualization techniques (such as
29-6 Vol. 3B
HANDLING BOUNDARY CONDITIONS IN A VIRTUAL MACHINE MONITOR
page-table virtualization) to trap guest accesses to the page frame hosting the
virtual local APIC registers.
Local APIC virtualization in the VMM needs to emulate the various local APIC operations and registers, such as: APIC identification/format registers, the local vector
table (LVT), the interrupt command register (ICR), interrupt capture registers (TMR,
IRR and ISR), task and processor priority registers (TPR, PPR), the EOI register and
the APIC-timer register. Since local APICs are designed to operate with non-specific
EOI, local APIC emulation also needs to emulate broadcast of EOI to the guest’s
virtual I/O APICs for level triggered virtual interrupts.
A local APIC allows interrupt masking at two levels: (1) mask bit in the local vector
table entry for local interrupts and (2) raising processor priority through the TPR
registers for masking lower priority external interrupts. The VMM needs to comprehend these virtual local APIC mask settings as programmed by the guest in addition
to the guest virtual processor interruptibility state (when injecting APIC routed
external virtual interrupts to a guest VM).
VMX provides several features which help the VMM to virtualize the local APIC. These
features allow many of guest TPR accesses (using CR8 only) to occur without VM
exits to the VMM:
•
The VMCS contains a “virtual-APIC address” field. This 64-bit field is the physical
address of the 4-KByte virtual APIC page (4-KByte aligned). The virtual-APIC
page contains a TPR shadow, which is accessed by the MOV CR8 instruction. The
TPR shadow comprises bits 7:4 in byte 80H of the virtual-APIC page.
•
The TPR threshold: bits 3:0 of this 32-bit field determine the threshold below
which the TPR shadow cannot fall. A VM exit will occur after an execution of MOV
CR8 that reduces the TPR shadow below this value.
•
The processor-based VM-execution controls field contains a “use TPR shadow” bit
and a “CR8-store exiting” bit. If the “use TPR shadow” VM-execution control is 1
and the “CR8-store exiting” VM-execution control is 0, then a MOV from CR8
reads from the TPR shadow. If the “CR8-store exiting” VM-execution control is 1,
then MOV from CR8 causes a VM exit; the “use TPR shadow” VM-execution
control is ignored in this case.
•
The processor-based VM-execution controls field contains a “CR8-load exiting”
bit. If the “use TPR shadow” VM-execution control is set and the “CR8-load
exiting” VM-execution control is clear, then MOV to CR8 writes to the “TPR
shadow”. A VM exit will occur after this write if the value written is below the TPR
threshold. If the “CR8-load exiting” VM-execution control is set, then MOV to CR8
causes a VM exit; the “use TPR shadow” VM-execution control is ignored in this
case.
29.3.2.4
I/O APIC Virtualization
The I/O APIC registers are typically mapped to a 1 MByte region where each I/O APIC
is allocated a 4K address window within this range. The VMM may utilize physical
memory virtualization to trap guest accesses to the virtual I/O APIC memory-
Vol. 3B 29-7
HANDLING BOUNDARY CONDITIONS IN A VIRTUAL MACHINE MONITOR
mapped registers. The I/O APIC virtualization needs to emulate the various I/O APIC
operations and registers such as identification/version registers, indirect-I/O-access
registers, EOI register, and the I/O redirection table. I/O APIC virtualization also
need to emulate various redirection table entry settings such as delivery mode,
destination mode, delivery status, polarity, masking, and trigger mode programmed
by the guest and track remote-IRR state on guest EOI writes to various virtual local
APICs.
29.3.2.5
Virtualization of Message Signaled Interrupts
The PCI Local Bus Specification (Rev. 2.2) introduces the concept of message
signaled interrupts (MSI). MSI enable PCI devices to request service by writing a
system-specified message to a system specified address. The transaction address
specifies the message destination while the transaction data specifies the interrupt
vector, trigger mode and delivery mode. System software is expected to configure
the message data and address during MSI device configuration, allocating one or
more no-shared messages to MSI capable devices. Chapter 10, “Advanced Programmable Interrupt Controller (APIC),” specifies the MSI message address and data
register formats to be followed on Intel 64 and IA-32 platforms. While MSI is optional
for conventional PCI devices, it is the preferred interrupt mechanism for PCI-Express
devices.
Since the MSI address and data are configured through PCI configuration space, to
control these physical interrupts the VMM needs to assume ownership of PCI configuration space. This allows the VMM to capture the guest configuration of message
address and data for MSI-capable virtual and assigned guest devices. PCI configuration transactions on PC-compatible systems are generated by software through two
different methods:
1. The standard CONFIG_ADDRESS/CONFIG_DATA register mechanism
(CFCH/CF8H ports) as defined in the PCI Local Bus Specification.
2. The enhanced flat memory-mapped (MEMCFG) configuration mechanism as
defined in the PCI-Express Base Specification (Rev. 1.0a.).
The CFCH/CF8H configuration access from guests can be trapped by the VMM
through use of I/O-bitmap VM-execution controls. The memory-mapped PCI-Express
MEMCFG guest configuration accesses can be trapped by VMM through physical
memory virtualization.
29.3.3
Examples of Handling of External Interrupts
The following sections illustrate interrupt processing in a VMM (when used to support
the external interrupt virtualization requirements).
29-8 Vol. 3B
HANDLING BOUNDARY CONDITIONS IN A VIRTUAL MACHINE MONITOR
29.3.3.1
Guest Setup
The VMM sets up the guest to cause a VM exit to the VMM on external interrupts. This
is done by setting the “external-interrupt exiting” VM-execution control in the guest
controlling-VMCS.
29.3.3.2
Processor Treatment of External Interrupt
Interrupts are automatically masked by hardware in the processor on VM exit by
clearing RFLAGS.IF. The exit-reason field in VMCS is set to 1 to indicate an external
interrupt as the exit reason.
If the VMM is utilizing the acknowledge-on-exit feature (by setting the “acknowledge
interrupt on exit” VM-exit control), the processor acknowledges the interrupt,
retrieves the host vector, and saves the interrupt in the VM-exit-interruption-information field (in the VM-exit information region of the VMCS) before transitioning
control to the VMM.
29.3.3.3
Processing of External Interrupts by VMM
Upon VM exit, the VMM can determine the exit cause of an external interrupt by
checking the exit-reason field (value = 1) in VMCS. If the acknowledge-interrupt-onexit control (see Section 21.7.1) is enabled, the VMM can use the saved host vector
(in the exit-interruption-information field) to switch to the appropriate interrupt
handler. If the “acknowledge interrupt on exit” VM-exit control is 0, the VMM may reenable interrupts (by setting RFLAGS.IF) to allow vectoring of external interrupts
through the monitor/host IDT.
The following steps may need to be performed by the VMM to process an external
interrupt:
•
Host Owned I/O Devices: For host-owned I/O devices, the interrupting device
is owned by the VMM (or hosting OS in a hosted VMM). In this model, the
interrupt service routine in the VMM/host driver is invoked and, upon ISR
completion, the appropriate write sequences (TPR updates, EOI etc.) to
respective interrupt controllers are performed as normal. If the work completion
indicated by the driver implies virtual device activity, the VMM runs the virtual
device emulation. Depending on the device class, physical device activity could
imply activity by multiple virtual devices mapped over the device. For each
affected virtual device, the VMM injects a virtual external interrupt event to
respective guest virtual machines. The guest driver interacts with the emulated
virtual device to process the virtual interrupt. The interrupt controller emulation
in the VMM supports various guest accesses to the VMM’s virtual interrupt
controller.
•
Guest Assigned I/O Devices: For assigned I/O devices, either the VMM uses a
software proxy or it can directly map the physical device to the assigned VM. In
both cases, servicing of the interrupt condition on the physical device is initiated
by the driver running inside the guest VM. With host control of external
Vol. 3B 29-9
HANDLING BOUNDARY CONDITIONS IN A VIRTUAL MACHINE MONITOR
interrupts, interrupts from assigned physical devices cause VM exits to the VMM
and vectoring through the host IDT to the registered VMM interrupt handler. To
unblock delivery of other low priority platform interrupts, the VMM interrupt
handler must mask the interrupt source (for level triggered interrupts) and issue
the appropriate EOI write sequences.
Once the physical interrupt source is masked and the platform EOI generated, the
VMM can map the host vector to its corresponding guest vector to inject the virtual
interrupt into the assigned VM. The guest software does EOI write sequences to its
virtual interrupt controller after completing interrupt processing. For level triggered
interrupts, these EOI writes to the virtual interrupt controller may be trapped by the
VMM which may in turn unmask the previously masked interrupt source.
29.3.3.4
Generation of Virtual Interrupt Events by VMM
The following provides some of the general steps that need to be taken by VMM
designs when generating virtual interrupts:
1. Check virtual processor interruptibility state. The virtual processor interruptibility
state is reflected in the guest RFLAGS.IF flag and the processor interruptibilitystate saved in the guest state area of the controlling-VMCS. If RFLAGS.IF is set
and the interruptibility state indicates readiness to take external interrupts (STImasking and MOV-SS/POP-SS-masking bits are clear), the guest virtual
processor is ready to take external interrupts. If the VMM design supports nonactive guest sleep states, the VMM needs to make sure the current guest sleep
state allows injection of external interrupt events.
2. If the guest virtual processor state is currently not interruptible, a VMM may
utilize the “interrupt-window exiting” VM-execution to notify the VM (through a
VM exit) when the virtual processor state changes to interruptible state.
3. Check the virtual interrupt controller state. If the guest VM exposes a virtual local
APIC, the current value of its processor priority register specifies if guest
software allows dispensing an external virtual interrupt with a specific priority to
the virtual processor. If the virtual interrupt is routed through the local vector
table (LVT) entry of the local APIC, the mask bits in the corresponding LVT entry
specifies if the interrupt is currently masked. Similarly, the virtual interrupt
controller’s current mask (IO-APIC or PIC) and priority settings reflect guest
state to accept specific external interrupts. The VMM needs to check both the
virtual processor and interrupt controller states to verify its guest interruptibility
state. If the guest is currently interruptible, the VMM can inject the virtual
interrupt. If the current guest state does not allow injecting a virtual interrupt,
the interrupt needs to be queued by the VMM until it can be delivered.
4. Prioritize the use of VM-entry event injection. A VMM may use VM-entry event
injection to deliver various virtual events (such as external interrupts,
exceptions, traps, and so forth). VMM designs may prioritize use of virtualinterrupt injection between these event types. Since each VM entry allows
injection of one event, depending on the VMM event priority policies, the VMM
may need to queue the external virtual interrupt if a higher priority event is to be
29-10 Vol. 3B
HANDLING BOUNDARY CONDITIONS IN A VIRTUAL MACHINE MONITOR
delivered on the next VM entry. Since the VMM has masked this particular
interrupt source (if it was level triggered) and done EOI to the platform interrupt
controller, other platform interrupts can be serviced while this virtual interrupt
event is queued for later delivery to the VM.
5. Update the virtual interrupt controller state. When the above checks have
passed, before generating the virtual interrupt to the guest, the VMM updates the
virtual interrupt controller state (Local-APIC, IO-APIC and/or PIC) to reflect
assertion of the virtual interrupt. This involves updating the various interrupt
capture registers, and priority registers as done by the respective hardware
interrupt controllers. Updating the virtual interrupt controller state is required for
proper interrupt event processing by guest software.
6. Inject the virtual interrupt on VM entry. To inject an external virtual interrupt to a
guest VM, the VMM sets up the VM-entry interruption-information field in the
guest controlling-VMCS before entry to guest using VMRESUME. Upon VM entry,
the processor will use this vector to access the gate in guest’s IDT and the value
of RFLAGS and EIP in guest-state area of controlling-VMCS is pushed on the
guest stack. If the guest RFLAGS.IF is clear, the STI-masking bit is set, or the
MOV- SS/POP-SS-masking bit is set, the VM entry will fail and the processor will
load state from the host-state area of the working VMCS as if a VM exit had
occurred (see Section 23.7).
29.4
ERROR HANDLING BY VMM
Error conditions may occur during VM entries and VM exits and a few other situations. This section describes how VMM should handle these error conditions,
including triple faults and machine check exceptions.
29.4.1
VM-Exit Failures
All VM exits load processor state from the host-state area of the VMCS that was the
controlling VMCS before the VM exit. This state is checked for consistency while being
loaded. Because the host-state is checked on VM entry, these checks will generally
succeed. Failure is possible only if host software is incorrect or if VMCS data in the
VMCS region in memory has been written by guest software (or by I/O DMA) since
the last VM entry. VM exits may fail for the following reasons:
•
•
•
There was a failure on storing guest MSRs.
•
•
There was a failure on loading host MSRs.
There was failure in loading a PDPTR.
The controlling VMCS has been corrupted (through writes to the corresponding
VMCS region) in such a way that the implementation cannot complete the VM
exit.
A machine check occurred.
Vol. 3B 29-11
HANDLING BOUNDARY CONDITIONS IN A VIRTUAL MACHINE MONITOR
If one of these problems occurs on a VM exit, a VMX abort results.
29.4.2
Machine Check Considerations
The following sequence determine how machine check exceptions are handled during
VMXON, VMXOFF, VM entries, and VM exits:
•
VMXOFF and VMXON:
If a machine check occurs during VMXOFF or VMXON and CR4.MCE = 1, a
machine-check exception (#MC) is generated. If CR4.MCE = 0, the processor
goes to shutdown state.
•
VM entry:
If a machine check occurs during VM entry, one of the following two treatments
must occur:
a. Normal delivery. If CR4.MCE = 1, delivery of a machine-check exception
(#MC) through the host IDT occurs. If CR4.MCE = 0, the processor goes to
shutdown state.
b. Load state from the host-state area of the working VMCS as if a VM exit had
occurred (see Section 23.7). The basic exit reason will be “VM-entry failure
due to machine check.”
If the machine check occurs after any guest state has been loaded, option b
above must be used. If the machine check occurs while checking host state and
VMX controls (or while reporting a failure due to such checks), option a should be
preferred; however, an implementation may use b, since software will not be able
to tell whether any guest state has been loaded.
•
VM exit:
If a machine check occurs during VM exit, one of the following two treatments
must occur:
— Normal delivery. If CR4.MCE = 1, delivery of a machine-check exception
(#MC) through the guest IDT. If CR4.MCE = 0, the processor goes to
shutdown state.
— Fail the VM exit. If the VM exit is to VMX root operation, a VMX abort will
result; it will block events as done normally in VMX abort. The VMX abort
indicator will show a machine check has induced the abort operation.
If a machine check is induced by an action in VMX non-root operation before any
determination is made that the inducing action may cause a VM exit, that
machine check should be considered as happening during guest execution in VMX
non-root operation. This is the case even if the part of the action that caused the
machine check was VMX-specific (for example: the processor’s consulting an I/O
bitmap). A machine-check exception will occur. If bit 12H of the exception bitmap
is cleared to 0, a machine-check exception could be delivered to the guest
through gate 12H of its IDT; if the bit is set to 1, the machine-check exception will
cause a VM exit.
29-12 Vol. 3B
HANDLING BOUNDARY CONDITIONS IN A VIRTUAL MACHINE MONITOR
NOTE
The state saved in the guest-state area on VM exits due to machinecheck exceptions should be considered suspect. A VMM should
consult the RIPV and EIPV bits in the IA32_MCG_STATUS MSR before
resuming a guest that caused a VM exit due to a machine-check
exception.
29.4.3
MCA Error Handling Guidelines for VMM
Section 29.4.2 covers general requirements for VMMs to handle machine-check
exceptions, when normal operation of the guest machine and/or the VMM is no
longer possible. enhancements of machine check architecture in newer processors
may support software recovery of uncorrected MC errors (UCR) signaled through
either machine-check exceptions or corrected machine-check interrupt (CMCI).
Section 15.5 and Section 15.6 describes details of these more recent enhancements
of machine check architecture.
In general, Virtual Machine Monitor (VMM) error handling should follow the recommendations for OS error handling described in Section 15.3, Section 15.6, Section
15.9, and Section 15.10. This section describes additional guidelines for hosted and
native hypervisor-based VMM implementations to support corrected MC errors and
recoverable uncorrected MC errors.
Because a hosted VMM provides virtualization services in the context of an existing
standard host OS, the host OS controls platform hardware through the host OS
services such as the standard OS device drivers. In hosted VMMs. MCA errors will be
handled by the host OS error handling software.
In native VMMs, the hypervisor runs on the hardware directly, and may provide only
a limited set of platform services for guest VMs. Most platform services may instead
be provided by a “control OS”. In hypervisor-based VMMs, MCA errors will either be
delivered directly to the VMM MCA handler (when the error is signaled while in the
VMM context) or cause by a VM exit from a guest VM or be delivered to the MCA intercept handler. There are two general approaches the hypervisor can use to handle the
MCA error: either within the hypervisor itself or by forwarding the error to the control
OS.
29.4.3.1
VMM Error Handling Strategies
Broadly speaking, there are two strategies that VMMs may take for error handling:
•
Basic error handling: in this approach the guest VM is treated as any other thread
of execution. If the error recovery action does not support restarting the thread
after handling the error, the guest VM should be terminated.
•
MCA virtualization: in this approach, the VMM virtualizes the MCA events and
hardware. This enables the VMM to intercept MCA events and inject an MCA into
Vol. 3B 29-13
HANDLING BOUNDARY CONDITIONS IN A VIRTUAL MACHINE MONITOR
the guest VM. The guest VM then has the opportunity to attempt error recovery
actions, rather than being terminated by the VMM.
Details of these approaches and implementation considerations for hosted and native
VMMs are discussed below.
29.4.3.2
Basic VMM MCA error recovery handling
The simplest approach is for the VMM to treat the guest VM as any other thread of
execution:
•
MCE's that occur outside the stream of execution of a virtual machine guest will
cause an MCE abort and may be handled by the MCA error handler following the
recovery actions and guidelines described in Section 15.9, and Section 15.10.
This includes logging the error and taking appropriate recovery actions when
necessary. The VMM must not resume the interrupted thread of execution or
another VM until it has taken the appropriate recovery action or, in the case of
fatal MCAs, reset the system.
•
MCE's that occur while executing in the context of a virtual machine will be
intercepted by the VMM. The MCA intercept handler may follow the error handling
guidelines listed in Section 15.9 and Section 15.10 for SRAO and SRAR errors.
For SRAR errors, terminating the thread of execution will involve terminating the
affected guest VM. For fatal errors the MCA handler should log the error and reset
the system -- the VMM should not resume execution of the interrupted VM.
29.4.3.3
Implementation Considerations for the Basic Model
For hosted VMMs, the host OS MCA error handling code will perform error analysis
and initiate the appropriate recovery actions. For the basic model this flow does not
change when terminating a guest VM although the specific actions needed to terminate a guest VM may be different than terminating an application or user process.
For native, hypervisor-based VMMs, MCA errors will either be delivered directly to the
VMM MCA handler (when the error is signaled while in the VMM context) or cause a
VM exit from a guest VM or be delivered to the MCA intercept handler. There are two
general approaches the hypervisor can use to handle the MCA error: either by
forwarding the error to the control OS or within the hypervisor itself. These
approaches are described in the following paragraphs.
The hypervisor may forward the error to the control OS for handling errors. This
approach simplifies the hypervisor error handling since it relies on the control OS to
implement the basic error handling model. The control OS error handling code will be
similar to the error handling code in the hosted VMM. Errors can be forwarded to the
control OS via an OS callback or by injecting an MCE event into the control OS.
Injecting an MCE will cause the control OS MCA error handler to be invoked. The
control OS is responsible for terminating the affected guest VM, if necessary, which
may require cooperation from the hypervisor.
29-14 Vol. 3B
HANDLING BOUNDARY CONDITIONS IN A VIRTUAL MACHINE MONITOR
Alternatively, the error may be handled completely in the hypervisor. The hypervisor
error handler is enhanced to implement the basic error handling model and the
hypervisor error handler has the capability to fully analyze the error information and
take recovery actions based on the guidelines. In this case error handling steps in the
hypervisor are similar to those for the hosted VMM described above (where the
hypervisor replaces the host OS actions). The hypervisor is responsible for terminating the affected guest VM, if necessary.
In all cases, if a fatal error is detected the VMM error handler should log the error and
reset the system. The VMM error handler must ensure that guest VMs are not
resumed after a fatal error is detected to ensure error containment is maintained.
29.4.3.4
MCA Virtualization
A more sophisticated approach for handling errors is to virtualize the MCA. This
involves virtualizing the MCA hardware and intercepting the MCA event in the VMM
when a guest VM is interrupted by an MCA. After analyzing the error, the VMM error
handler may then decide to inject an MCE abort into the guest VM for attempted
guest VM error recovery. This would enable the guest OS the opportunity to take
recovery actions specific to that guest.
For MCA virtualization, the VMM must provide the guest physical address for memory
errors instead of the system physical address when reporting the errors to the guest
VM. To compute the guest physical address, the VMM needs to maintain a reverse
mapping of system physical page addresses to guest physical page addresses.
When the MCE is injected into the guest VM, the guest OS MCA handler would be
invoked. The guest OS implements the MCA handling guidelines and it could potentially terminate the interrupted thread of execution within the guest instead of terminating the VM. The guest OS may also disable use of the affected page by the guest.
When disabling the page the VMM error handler may handle the case where a page is
shared by the VMM and a guest or by two guests. In these cases the page use must
be disabled in both contexts to ensure no subsequent consumption errors are generated.
29.4.3.5
Implementation Considerations for the MCA Virtualization Model
MCA virtualization may be done in either hosted VMMs or hypervisor-based VMMs.
The error handling flow is similar to the flow described in the basic handling case. The
major difference is that the recovery action includes injecting the MCE abort into the
guest VM to enable recovery by the guest OS when the MCA interrupts the execution
of a guest VM.
29.5
HANDLING ACTIVITY STATES BY VMM
A VMM might place a logic processor in the wait-for-SIPI activity state if supporting
certain guest operating system using the multi-processor (MP) start-up algorithm. A
Vol. 3B 29-15
HANDLING BOUNDARY CONDITIONS IN A VIRTUAL MACHINE MONITOR
guest with direct access to the physical local APIC and using the MP start-up algorithm sends an INIT-SIPI-SIPI IPI sequence to start the application processor. In
order to trap the SIPIs, the VMM must start the logic processor which is the target of
the SIPIs in wait-for-SIPI mode.
29-16 Vol. 3B
CHAPTER 30
PERFORMANCE MONITORING
Intel 64 and IA-32 architectures provide facilities for monitoring performance.
30.1
PERFORMANCE MONITORING OVERVIEW
Performance monitoring was introduced in the Pentium processor with a set of
model-specific performance-monitoring counter MSRs. These counters permit selection of processor performance parameters to be monitored and measured. The information obtained from these counters can be used for tuning system and compiler
performance.
In Intel P6 family of processors, the performance monitoring mechanism was
enhanced to permit a wider selection of events to be monitored and to allow greater
control events to be monitored. Next, Pentium 4 and Intel Xeon processors introduced a new performance monitoring mechanism and new set of performance
events.
The performance monitoring mechanisms and performance events defined for the
Pentium, P6 family, Pentium 4, and Intel Xeon processors are not architectural. They
are all model specific (not compatible among processor families). Intel Core Solo and
Intel Core Duo processors support a set of architectural performance events and a
set of non-architectural performance events. Processors based on Intel Core
microarchitecture and Intel® Atom™ microarchitecture support enhanced architectural performance events and non-architectural performance events.
Starting with Intel Core Solo and Intel Core Duo processors, there are two classes of
performance monitoring capabilities. The first class supports events for monitoring
performance using counting or sampling usage. These events are non-architectural
and vary from one processor model to another. They are similar to those available in
Pentium M processors. These non-architectural performance monitoring events are
specific to the microarchitecture and may change with enhancements. They are
discussed in Section 30.3, “Performance Monitoring (Intel® Core™ Solo and Intel®
Core™ Duo Processors).” Non-architectural events for a given microarchitecture can
not be enumerated using CPUID; and they are listed in Appendix A, “PerformanceMonitoring Events.”
The second class of performance monitoring capabilities is referred to as architectural performance monitoring. This class supports the same counting and sampling
usages, with a smaller set of available events. The visible behavior of architectural
performance events is consistent across processor implementations. Availability of
architectural performance monitoring capabilities is enumerated using the
CPUID.0AH. These events are discussed in Section 30.2.
See also:
Vol. 3B 30-1
PERFORMANCE MONITORING
— Section 30.2, “Architectural Performance Monitoring”
— Section 30.3, “Performance Monitoring (Intel® Core™ Solo and Intel® Core™
Duo Processors)”
— Section 30.4, “Performance Monitoring (Processors Based on Intel® Core™
Microarchitecture)”
— Section 30.5, “Performance Monitoring (Processors Based on Intel® Atom™
Microarchitecture)”
— Section 30.6, “Performance Monitoring for Processors Based on Intel®
Microarchitecture Code Name Nehalem”
— Section 30.7, “Performance Monitoring for Processors Based on Intel®
Microarchitecture Code Name Westmere”
— Section 30.8, “Performance Monitoring for Processors Based on Intel®
Microarchitecture Code Name Sandy Bridge”
— Section 30.9, “Performance Monitoring (Processors Based on Intel NetBurst®
Microarchitecture)”
— Section 30.10, “Performance Monitoring and Intel Hyper-Threading
Technology in Processors Based on Intel NetBurst® Microarchitecture”
— Section 30.13, “Performance Monitoring and Dual-Core Technology”
— Section 30.14, “Performance Monitoring on 64-bit Intel Xeon Processor MP
with Up to 8-MByte L3 Cache”
— Section 30.16, “Performance Monitoring (P6 Family Processor)”
— Section 30.17, “Performance Monitoring (Pentium Processors)”
30.2
ARCHITECTURAL PERFORMANCE MONITORING
Performance monitoring events are architectural when they behave consistently
across microarchitectures. Intel Core Solo and Intel Core Duo processors introduced
architectural performance monitoring. The feature provides a mechanism for software to enumerate performance events and provides configuration and counting
facilities for events.
Architectural performance monitoring does allow for enhancement across processor
implementations. The CPUID.0AH leaf provides version ID for each enhancement.
Intel Core Solo and Intel Core Duo processors support base level functionality identified by version ID of 1. Processors based on Intel Core microarchitecture support, at
a minimum, the base level functionality of architectural performance monitoring.
Intel Core 2 Duo processor T 7700 and newer processors based on Intel Core
microarchitecture support both the base level functionality and enhanced architectural performance monitoring identified by version ID of 2.
30-2 Vol. 3B
PERFORMANCE MONITORING
Intel Atom processor family supports the base level functionality, enhanced architectural performance monitoring identified by version ID of 2 and version ID of 3
(including two general-purpose performance counters, IA32_PMC0, IA32_PMC1).
Intel Core i7 processor family supports the base level functionality, enhanced architectural performance monitoring identified by version ID of 2 and version ID of 3,
(including four general-purpose performance counters, IA32_PMC0-IA32_PMC3).
30.2.1
Architectural Performance Monitoring Version 1
Configuring an architectural performance monitoring event involves programming
performance event select registers. There are a finite number of performance event
select MSRs (IA32_PERFEVTSELx MSRs). The result of a performance monitoring
event is reported in a performance monitoring counter (IA32_PMCx MSR). Performance monitoring counters are paired with performance monitoring select registers.
Performance monitoring select registers and counters are architectural in the
following respects:
•
•
Bit field layout of IA32_PERFEVTSELx is consistent across microarchitectures.
•
•
Addresses of IA32_PMC MSRs remain the same across microarchitectures.
Addresses of IA32_PERFEVTSELx MSRs remain the same across microarchitectures.
Each logical processor has its own set of IA32_PERFEVTSELx and IA32_PMCx
MSRs. Configuration facilities and counters are not shared between logical
processors sharing a processor core.
Architectural performance monitoring provides a CPUID mechanism for enumerating
the following information:
•
Number of performance monitoring counters available in a logical processor
(each IA32_PERFEVTSELx MSR is paired to the corresponding IA32_PMCx MSR)
•
•
Number of bits supported in each IA32_PMCx
Number of architectural performance monitoring events supported in a logical
processor
Software can use CPUID to discover architectural performance monitoring availability
(CPUID.0AH). The architectural performance monitoring leaf provides an identifier
corresponding to the version number of architectural performance monitoring available in the processor.
The version identifier is retrieved by querying CPUID.0AH:EAX[bits 7:0] (see
Chapter 3, “Instruction Set Reference, A-M,” in the Intel® 64 and IA-32 Architectures
Software Developer’s Manual, Volume 2A). If the version identifier is greater than
zero, architectural performance monitoring capability is supported. Software queries
the CPUID.0AH for the version identifier first; it then analyzes the value returned in
CPUID.0AH.EAX, CPUID.0AH.EBX to determine the facilities available.
In the initial implementation of architectural performance monitoring; software can
determine how many IA32_PERFEVTSELx/ IA32_PMCx MSR pairs are supported per
Vol. 3B 30-3
PERFORMANCE MONITORING
core, the bit-width of PMC, and the number of architectural performance monitoring
events available.
30.2.1.1
Architectural Performance Monitoring Version 1 Facilities
Architectural performance monitoring facilities include a set of performance monitoring counters and performance event select registers. These MSRs have the
following properties:
•
IA32_PMCx MSRs start at address 0C1H and occupy a contiguous block of MSR
address space; the number of MSRs per logical processor is reported using
CPUID.0AH:EAX[15:8].
•
IA32_PERFEVTSELx MSRs start at address 186H and occupy a contiguous block
of MSR address space. Each performance event select register is paired with a
corresponding performance counter in the 0C1H address block.
•
The bit width of an IA32_PMCx MSR is reported using the
CPUID.0AH:EAX[23:16]. This the number of valid bits for read operation. On
write operations, the lower-order 32 bits of the MSR may be written with any
value, and the high-order bits are sign-extended from the value of bit 31.
•
Bit field layout of IA32_PERFEVTSELx MSRs is defined architecturally.
See Figure 30-1 for the bit field layout of IA32_PERFEVTSELx MSRs. The bit fields
are:
•
Event select field (bits 0 through 7) — Selects the event logic unit used to
detect microarchitectural conditions (see Table 30-1, for a list of architectural
events and their 8-bit codes). The set of values for this field is defined architecturally; each value corresponds to an event logic unit for use with an architectural
performance event. The number of architectural events is queried using
CPUID.0AH:EAX. A processor may support only a subset of pre-defined values.
30-4 Vol. 3B
PERFORMANCE MONITORING
63
31
24 23 22 21 20 19 18 17 16 15
Counter Mask I E
N
(CMASK)
V N
0
8 7
I
U
N P E O S Unit Mask (UMASK)
S R
T C
Event Select
INV—Invert counter mask
EN—Enable counters
INT—APIC interrupt enable
PC—Pin control
E—Edge detect
OS—Operating system mode
USR—User Mode
Reserved
Figure 30-1. Layout of IA32_PERFEVTSELx MSRs
•
Unit mask (UMASK) field (bits 8 through 15) — These bits qualify the
condition that the selected event logic unit detects. Valid UMASK values for each
event logic unit are specific to the unit. For each architectural performance event,
its corresponding UMASK value defines a specific microarchitectural condition.
A pre-defined microarchitectural condition associated with an architectural event
may not be applicable to a given processor. The processor then reports only a
subset of pre-defined architectural events. Pre-defined architectural events are
listed in Table 30-1; support for pre-defined architectural events is enumerated
using CPUID.0AH:EBX. Architectural performance events available in the initial
implementation are listed in Table A-1.
•
USR (user mode) flag (bit 16) — Specifies that the selected microarchitectural
condition is counted only when the logical processor is operating at privilege
levels 1, 2 or 3. This flag can be used with the OS flag.
•
OS (operating system mode) flag (bit 17) — Specifies that the selected
microarchitectural condition is counted only when the logical processor is
operating at privilege level 0. This flag can be used with the USR flag.
•
E (edge detect) flag (bit 18) — Enables (when set) edge detection of the
selected microarchitectural condition. The logical processor counts the number of
deasserted to asserted transitions for any condition that can be expressed by the
other fields. The mechanism does not permit back-to-back assertions to be
distinguished.
This mechanism allows software to measure not only the fraction of time spent in
a particular state, but also the average length of time spent in such a state (for
example, the time spent waiting for an interrupt to be serviced).
Vol. 3B 30-5
PERFORMANCE MONITORING
•
PC (pin control) flag (bit 19) — When set, the logical processor toggles the
PMi pins and increments the counter when performance-monitoring events
occur; when clear, the processor toggles the PMi pins when the counter
overflows. The toggling of a pin is defined as assertion of the pin for a single bus
clock followed by deassertion.
•
INT (APIC interrupt enable) flag (bit 20) — When set, the logical processor
generates an exception through its local APIC on counter overflow.
•
EN (Enable Counters) Flag (bit 22) — When set, performance counting is
enabled in the corresponding performance-monitoring counter; when clear, the
corresponding counter is disabled. The event logic unit for a UMASK must be
disabled by setting IA32_PERFEVTSELx[bit 22] = 0, before writing to
IA32_PMCx.
•
INV (invert) flag (bit 23) — Inverts the result of the counter-mask comparison
when set, so that both greater than and less than comparisons can be made.
•
Counter mask (CMASK) field (bits 24 through 31) — When this field is not
zero, a logical processor compares this mask to the events count of the detected
microarchitectural condition during a single cycle. If the event count is greater
than or equal to this mask, the counter is incremented by one. Otherwise the
counter is not incremented.
This mask is intended for software to characterize microarchitectural conditions
that can count multiple occurrences per cycle (for example, two or more instructions retired per clock; or bus queue occupations). If the counter-mask field is 0,
then the counter is incremented each cycle by the event count associated with
multiple occurrences.
30.2.2
Additional Architectural Performance Monitoring Extensions
The enhanced features provided by architectural performance monitoring version 2
include the following:
•
Fixed-function performance counter register and associated control
register — Three of the architectural performance events are counted using
three fixed-function MSRs (IA32_FIXED_CTR0 through IA32_FIXED_CTR2). Each
of the fixed-function PMC can count only one architectural performance event.
Configuring the fixed-function PMCs is done by writing to bit fields in the MSR
(IA32_FIXED_CTR_CTRL) located at address 38DH. Unlike configuring
performance events for general-purpose PMCs (IA32_PMCx) via UMASK field in
(IA32_PERFEVTSELx), configuring, programming IA32_FIXED_CTR_CTRL for
fixed-function PMCs do not require any UMASK.
•
Simplified event programming — Most frequent operation in programming
performance events are enabling/disabling event counting and checking the
status of counter overflows. Architectural performance event version 2 provides
three architectural MSRs:
30-6 Vol. 3B
PERFORMANCE MONITORING
— IA32_PERF_GLOBAL_CTRL allows software to enable/disable event counting
of all or any combination of fixed-function PMCs (IA32_FIXED_CTRx) or any
general-purpose PMCs via a single WRMSR.
— IA32_PERF_GLOBAL_STATUS allows software to query counter overflow
conditions on any combination of fixed-function PMCs or general-purpose
PMCs via a single RDMSR.
— IA32_PERF_GLOBAL_OVF_CTRL allows software to clear counter overflow
conditions on any combination of fixed-function PMCs or general-purpose
PMCs via a single WRMSR.
30.2.2.1
Architectural Performance Monitoring Version 2 Facilities
The facilities provided by architectural performance monitoring version 2 can be
queried from CPUID leaf 0AH by examining the content of register EDX:
•
Bits 0 through 4 of CPUID.0AH.EDX indicates the number of fixed-function
performance counters available per core,
•
Bits 5 through 12 of CPUID.0AH.EDX indicates the bit-width of fixed-function
performance counters. Bits beyond the width of the fixed-function counter are
reserved and must be written as zeros.
NOTE
Early generation of processors based on Intel Core microarchitecture
may report in CPUID.0AH:EDX of support for version 2 but indicating
incorrect information of version 2 facilities.
The IA32_FIXED_CTR_CTRL MSR include multiple sets of 4-bit field, each 4 bit
field controls the operation of a fixed-function performance counter. Figure 30-2
shows the layout of 4-bit controls for each fixed-function PMC. Two sub-fields are
currently defined within each control. The definitions of the bit fields are:
63
12 11
P
M
I
9 8 7
E
N
P
M
I
5 43 2 1 0
E
N
P
M
I
E
N
Cntr2 — Controls for IA32_FIXED_CTR2
Cntr1 — Controls for IA32_FIXED_CTR1
PMI — Enable PMI on overflow
Cntr0 — Controls for IA32_FIXED_CTR0
ENABLE — 0: disable; 1: OS; 2: User; 3: All ring levels
Reserved
Figure 30-2. Layout of IA32_FIXED_CTR_CTRL MSR
Vol. 3B 30-7
PERFORMANCE MONITORING
•
Enable field (lowest 2 bits within each 4-bit control) — When bit 0 is set,
performance counting is enabled in the corresponding fixed-function
performance counter to increment while the target condition associated with the
architecture performance event occurred at ring 0. When bit 1 is set,
performance counting is enabled in the corresponding fixed-function
performance counter to increment while the target condition associated with the
architecture performance event occurred at ring greater than 0. Writing 0 to both
bits stops the performance counter. Writing a value of 11B enables the counter to
increment irrespective of privilege levels.
•
PMI field (the fourth bit within each 4-bit control) — When set, the logical
processor generates an exception through its local APIC on overflow condition of
the respective fixed-function counter.
IA32_PERF_GLOBAL_CTRL MSR provides single-bit controls to enable counting of
each performance counter. Figure 30-3 shows the layout of
IA32_PERF_GLOBAL_CTRL. Each enable bit in IA32_PERF_GLOBAL_CTRL is AND’ed
with the enable bits for all privilege levels in the respective IA32_PERFEVTSELx or
IA32_PERF_FIXED_CTR_CTRL MSRs to start/stop the counting of respective
counters. Counting is enabled if the AND’ed results is true; counting is disabled when
the result is false.
63
35 34 33 32 31
2 1 0
IA32_FIXED_CTR2 enable
IA32_FIXED_CTR1 enable
IA32_FIXED_CTR0 enable
IA32_PMC1 enable
IA32_PMC0 enable
Reserved
Figure 30-3. Layout of IA32_PERF_GLOBAL_CTRL MSR
The fixed-function performance counters supported by architectural performance
version 2 is listed in Table 30-8, the pairing between each fixed-function performance counter to an architectural performance event is also shown.
IA32_PERF_GLOBAL_STATUS MSR provides single-bit status for software to query
the overflow condition of each performance counter. The MSR also provides additional status bit to indicate overflow conditions when counters are programmed for
precise-event-based sampling (PEBS). IA32_PERF_GLOBAL_STATUS MSR also
provides a sticky bit to indicate changes to the state of performance monitoring hard-
30-8 Vol. 3B
PERFORMANCE MONITORING
ware. Figure 30-4 shows the layout of IA32_PERF_GLOBAL_STATUS. A value of 1 in
bits 0, 1, 32 through 34 indicates a counter overflow condition has occurred in the
associated counter.
When a performance counter is configured for PEBS, overflow condition in the
counter generates a performance-monitoring interrupt signaling a PEBS event. On a
PEBS event, the processor stores data records into the buffer area (see Section
18.15.5), clears the counter overflow status., and sets the “OvfBuffer” bit in
IA32_PERF_GLOBAL_STATUS.
63 62
35 34 33 32 31
2 1 0
CondChgd
OvfBuffer
IA32_FIXED_CTR2 Overflow
IA32_FIXED_CTR1 Overflow
IA32_FIXED_CTR0 Overflow
IA32_PMC1 Overflow
IA32_PMC0 Overflow
Reserved
Figure 30-4. Layout of IA32_PERF_GLOBAL_STATUS MSR
IA32_PERF_GLOBAL_OVF_CTL MSR allows software to clear overflow indicator(s) of
any general-purpose or fixed-function counters via a single WRMSR. Software should
clear overflow indications when
•
Setting up new values in the event select and/or UMASK field for counting or
sampling
•
•
Reloading counter values to continue sampling
Disabling event counting or sampling.
The layout of IA32_PERF_GLOBAL_OVF_CTL is shown in Figure 30-5.
Vol. 3B 30-9
PERFORMANCE MONITORING
63 62
35 34 33 32 31
2 1 0
ClrCondChgd
ClrOvfBuffer
IA32_FIXED_CTR2 ClrOverflow
IA32_FIXED_CTR1 ClrOverflow
IA32_FIXED_CTR0 ClrOverflow
IA32_PMC1 ClrOverflow
IA32_PMC0 ClrOverflow
Reserved
Figure 30-5. Layout of IA32_PERF_GLOBAL_OVF_CTRL MSR
30.2.2.2
Architectural Performance Monitoring Version 3 Facilities
The facilities provided by architectural performance monitoring version 1 and 2 are
also supported by architectural performance monitoring version 3. Additionally
version 3 provides enhancements to support a processor core comprising of more
than one logical processor, i.e. a processor core supporting Intel Hyper-Threading
Technology or simultaneous multi-threading capability. Specifically,
•
CPUID leaf 0AH provides enumeration mechanisms to query:
— The number of general-purpose performance counters (IA32_PMCx) is
reported in CPUID.0AH:EAX[15:8], the bit width of general-purpose
performance counters (see also Section 30.2.1.1) is reported in
CPUID.0AH:EAX[23:16].
— The bit vector representing the set of architectural performance monitoring
events supported (see Section 30.2.3)
— The number of fixed-function performance counters, the bit width of fixedfunction performance counters (see also Section 30.2.2.1).
•
Each general-purpose performance counter IA32_PMCx (starting at MSR address
0C1H) is associated with a corresponding IA32_PERFEVTSELx MSR (starting at
MSR address 186H). The Bit field layout of IA32_PERFEVTSELx MSRs is defined
architecturally in Figure 30-6.
30-10 Vol. 3B
PERFORMANCE MONITORING
63
31
24 23 22 21 20 19 18 17 16 15
0
8 7
A I
U
Counter Mask I E N
N
N P E O S Unit Mask (UMASK)
(CMASK)
S R
V N Y T C
Event Select
INV—Invert counter mask
EN—Enable counters
ANY—Any Thread
INT—APIC interrupt enable
PC—Pin control
E—Edge detect
OS—Operating system mode
USR—User Mode
Reserved
Figure 30-6. Layout of IA32_PERFEVTSELx MSRs Supporting Architectural
Performance Monitoring Version 3
Bit 21 (AnyThread) of IA32_PERFEVTSELx is supported in architectural
performance monitoring version 3. When set to 1, it enables counting the
associated event conditions (including matching the thread’s CPL with the
OS/USR setting of IA32_PERFEVTSELx) occurring across all logical processors
sharing a processor core. When bit 21 is 0, the counter only increments the
associated event conditions (including matching the thread’s CPL with the
OS/USR setting of IA32_PERFEVTSELx) occurring in the logical processor which
programmed the IA32_PERFEVTSELx MSR.
•
Each fixed-function performance counter IA32_FIXED_CTRx (starting at MSR
address 309H) is configured by a 4-bit control block in the
IA32_PERF_FIXED_CTR_CTRL MSR. The control block also allow threadspecificity configuration using an AnyThread bit. The layout of
IA32_PERF_FIXED_CTR_CTRL MSR is shown.
Vol. 3B 30-11
PERFORMANCE MONITORING
63
12 11
P A
M N
I Y
9 8 7
E
N
P A
M N
I Y
5 43 2 1 0
E
N
P A
M N
I Y
E
N
Cntr2 — Controls for IA32_FIXED_CTR2
Cntr1 — Controls for IA32_FIXED_CTR1
PMI — Enable PMI on overflow on IA32_FIXED_CTR0
AnyThread — AnyThread for IA32_FIXED_CTR0
ENABLE — IA32_FIXED_CTR0. 0: disable; 1: OS; 2: User; 3: All ring levels
Reserved
Figure 30-7. Layout of IA32_FIXED_CTR_CTRL MSR Supporting Architectural
Performance Monitoring Version 3
Each control block for a fixed-function performance counter provides a
AnyThread (bit position 2 + 4*N, N= 0, 1, etc.) bit. When set to 1, it enables
counting the associated event conditions (including matching the thread’s CPL
with the ENABLE setting of the corresponding control block of
IA32_PERF_FIXED_CTR_CTRL) occurring across all logical processors sharing a
processor core. When an AnyThread bit is 0 in IA32_PERF_FIXED_CTR_CTRL,
the corresponding fixed counter only increments the associated event conditions
occurring in the logical processor which programmed the
IA32_PERF_FIXED_CTR_CTRL MSR.
•
The IA32_PERF_GLOBAL_CTRL, IA32_PERF_GLOBAL_STATUS,
IA32_PERF_GLOBAL_OVF_CTRL MSRs provide single-bit controls/status for each
general-purpose and fixed-function performance counter. Figure 30-8 shows the
layout of these MSR for N general-purpose performance counters (where N is
reported by CPUID.0AH:EAX[15:8] ) and three fixed-function counters.
Note: Intel Atom processor family supports two general-purpose performance
monitoring counters (i.e. N =2 in Figure 30-8), other processor families in Intel
64 architecture may support a different value of N in Figure 30-8. The number N
is reported by CPUID.0AH:EAX[15:8]. Intel Core i7 processor family supports
four general-purpose performance monitoring counters (i.e. N =4 in Figure 30-8)
30-12 Vol. 3B
PERFORMANCE MONITORING
Global Enable Controls IA32_PERF_GLOBAL_CTRL
35 34 33 32 31
63
IA32_FIXED_CTR2 enable
IA32_FIXED_CTR1 enable
IA32_FIXED_CTR0 enable
IA32_PMC(N-1) enable
.................... enable
IA32_PMC1 enable
IA32_PMC0 enable
63 62
63 62
Reserved
Global Overflow Status IA32_PERF_GLOBAL_STATUS
35 34 33 32 31
CondChgd
OvfBuffer
IA32_FIXED_CTR2 Overflow
IA32_FIXED_CTR1 Overflow
IA32_FIXED_CTR0 Overflow
IA32_PMC1 Overflow
IA32_PMC0 Overflow
N .. .. 1 0
N .. .. 1 0
IA32_PMC(N-1) Overflow
...................... Overflow
Global Overflow Status IA32_PERF_GLOBAL_OVF_CTRL
35 34 33 32 31
N .. .. 1 0
ClrCondChgd
ClrOvfBuffer
IA32_FIXED_CTR2 ClrOverflow
IA32_FIXED_CTR1 ClrOverflow
IA32_FIXED_CTR0 ClrOverflow
IA32_PMC1 ClrOverflow
IA32_PMC0 ClrOverflow
IA32_PMC(N-1) ClrOverflow
........................ ClrOverflow
Figure 30-8. Layout of Global Performance Monitoring Control MSR
30.2.2.3
Full-Width Writes to Performance Counter Registers
The general-purpose performance counter registers IA32_PMCx are writable via
WRMSR instruction. However, the value written into IA32_PMCx by WRMSR is the
signed extended 64-bit value of the EAX[31:0] input of WRMSR.
A processor that supports full-width writes to the general-purpose performance
counters enumerated by CPUID.0AH:EAX[15:8] will set
Vol. 3B 30-13
PERFORMANCE MONITORING
IA32_PERF_CAPABILITIES[13] to enumerate its full-width-write capability See
Figure 30-39.
If IA32_PERF_CAPABILITIES.FW_WRITE[bit 13] =1, each IA32_PMCi is accompanied by a corresponding alias address starting at 4C1H for IA32_A_PMC0.
If IA32_A_PMCi is present, the 64-bit input value (EDX:EAX) of WRMSR to
IA32_A_PMCi will cause IA32_PMCi to be updated by:
IA32_PMCi[63:32] ← SignExtend(EDX[N-32:0]);
IA32_PMCi[31:0] ← EAX[31:0];
30.2.3
Pre-defined Architectural Performance Events
Table 30-1 lists architecturally defined events.
Table 30-1. UMask and Event Select Encodings for Pre-Defined
Architectural Performance Events
Bit Position
CPUID.AH.EBX
Event Name
UMask
Event Select
0
UnHalted Core Cycles
00H
3CH
1
Instruction Retired
00H
C0H
2
UnHalted Reference Cycles
01H
3CH
3
LLC Reference
4FH
2EH
4
LLC Misses
41H
2EH
5
Branch Instruction Retired
00H
C4H
6
Branch Misses Retired
00H
C5H
A processor that supports architectural performance monitoring may not support all
the predefined architectural performance events (Table 30-1). The non-zero bits in
CPUID.0AH:EBX indicate the events that are not available.
The behavior of each architectural performance event is expected to be consistent on
all processors that support that event. Minor variations between microarchitectures
are noted below:
•
UnHalted Core Cycles — Event select 3CH, Umask 00H
This event counts core clock cycles when the clock signal on a specific core is
running (not halted). The counter does not advance in the following conditions:
— an ACPI C-state other than C0 for normal operation
— HLT
— STPCLK# pin asserted
— being throttled by TM1
30-14 Vol. 3B
PERFORMANCE MONITORING
— during the frequency switching phase of a performance state transition (see
Chapter 14, “Power and Thermal Management”)
The performance counter for this event counts across performance state
transitions using different core clock frequencies
•
Instructions Retired — Event select C0H, Umask 00H
This event counts the number of instructions at retirement. For instructions that
consist of multiple micro-ops, this event counts the retirement of the last microop of the instruction. An instruction with a REP prefix counts as one instruction
(not per iteration). Faults before the retirement of the last micro-op of a multiops instruction are not counted.
This event does not increment under VM-exit conditions. Counters continue
counting during hardware interrupts, traps, and inside interrupt handlers.
•
UnHalted Reference Cycles — Event select 3CH, Umask 01H
This event counts reference clock cycles while the clock signal on the core is
running. The reference clock operates at a fixed frequency, irrespective of core
frequency changes due to performance state transitions. Processors may
implement this behavior differently. See Table A-10 and Table A-12 in Appendix
A, “Performance-Monitoring Events.”
•
Last Level Cache References — Event select 2EH, Umask 4FH
This event counts requests originating from the core that reference a cache line
in the last level cache. The event count includes speculation and cache line fills
due to the first-level cache hardware prefetcher, but may exclude cache line fills
due to other hardware-prefetchers.
Because cache hierarchy, cache sizes and other implementation-specific characteristics; value comparison to estimate performance differences is not recommended.
•
Last Level Cache Misses — Event select 2EH, Umask 41H
This event counts each cache miss condition for references to the last level cache.
The event count may include speculation and cache line fills due to the first-level
cache hardware prefetcher, but may exclude cache line fills due to other
hardware-prefetchers.
Because cache hierarchy, cache sizes and other implementation-specific characteristics; value comparison to estimate performance differences is not recommended.
•
Branch Instructions Retired — Event select C4H, Umask 00H
This event counts branch instructions at retirement. It counts the retirement of
the last micro-op of a branch instruction.
•
All Branch Mispredict Retired — Event select C5H, Umask 00H
This event counts mispredicted branch instructions at retirement. It counts the
retirement of the last micro-op of a branch instruction in the architectural path of
execution and experienced misprediction in the branch prediction hardware.
Vol. 3B 30-15
PERFORMANCE MONITORING
Branch prediction hardware is implementation-specific across microarchitectures; value comparison to estimate performance differences is not recommended.
NOTE
Programming decisions or software precisians on functionality should
not be based on the event values or dependent on the existence of
performance monitoring events.
30.3
PERFORMANCE MONITORING (INTEL® CORE™ SOLO
AND INTEL® CORE™ DUO PROCESSORS)
In Intel Core Solo and Intel Core Duo processors, non-architectural performance
monitoring events are programmed using the same facilities (see Figure 30-1) used
for architectural performance events.
Non-architectural performance events use event select values that are modelspecific. Event mask (Umask) values are also specific to event logic units. Some
microarchitectural conditions detectable by a Umask value may have specificity
related to processor topology (see Section 8.6, “Detecting Hardware Multi-Threading
Support and Topology,” in the Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 3A). As a result, the unit mask field (for example,
IA32_PERFEVTSELx[bits 15:8]) may contain sub-fields that specify topology information of processor cores.
The sub-field layout within the Umask field may support two-bit encoding that qualifies the relationship between a microarchitectural condition and the originating core.
This data is shown in Table 30-2. The two-bit encoding for core-specificity is only
supported for a subset of Umask values (see Appendix A, “Performance Monitoring
Events”) and for Intel Core Duo processors. Such events are referred to as corespecific events.
Table 30-2. Core Specificity Encoding within a Non-Architectural Umask
IA32_PERFEVTSELx MSRs
Bit 15:14 Encoding
Description
11B
All cores
10B
Reserved
01B
This core
00B
Reserved
Some microarchitectural conditions allow detection specificity only at the boundary
of physical processors. Some bus events belong to this category, providing specificity
between the originating physical processor (a bus agent) versus other agents on the
bus. Sub-field encoding for agent specificity is shown in Table 30-3.
30-16 Vol. 3B
PERFORMANCE MONITORING
Table 30-3. Agent Specificity Encoding within a Non-Architectural Umask
IA32_PERFEVTSELx MSRs
Bit 13 Encoding
Description
0
This agent
1
Include all agents
Some microarchitectural conditions are detectable only from the originating core. In
such cases, unit mask does not support core-specificity or agent-specificity encodings. These are referred to as core-only conditions.
Some microarchitectural conditions allow detection specificity that includes or
excludes the action of hardware prefetches. A two-bit encoding may be supported to
qualify hardware prefetch actions. Typically, this applies only to some L2 or bus
events. The sub-field encoding for hardware prefetch qualification is shown in
Table 30-4.
Table 30-4. HW Prefetch Qualification Encoding within a Non-Architectural Umask
IA32_PERFEVTSELx MSRs
Bit 13:12 Encoding
Description
11B
All inclusive
10B
Reserved
01B
Hardware prefetch only
00B
Exclude hardware prefetch
Some performance events may (a) support none of the three event-specific qualification encodings (b) may support core-specificity and agent specificity simultaneously
(c) or may support core-specificity and hardware prefetch qualification simultaneously. Agent-specificity and hardware prefetch qualification are mutually exclusive.
In addition, some L2 events permit qualifications that distinguish cache coherent
states. The sub-field definition for cache coherency state qualification is shown in
Table 30-5. If no bits in the MESI qualification sub-field are set for an event that
requires setting MESI qualification bits, the event count will not increment.
Table 30-5. MESI Qualification Definitions within a Non-Architectural Umask
IA32_PERFEVTSELx MSRs
Bit Position 11:8
Description
Bit 11
Counts modified state
Bit 10
Counts exclusive state
Vol. 3B 30-17
PERFORMANCE MONITORING
Table 30-5. MESI Qualification Definitions within a Non-Architectural Umask
IA32_PERFEVTSELx MSRs
Bit Position 11:8
Description
Bit 9
Counts shared state
Bit 8
Counts Invalid state
30.4
PERFORMANCE MONITORING (PROCESSORS BASED
ON INTEL® CORE™ MICROARCHITECTURE)
In addition to architectural performance monitoring, processors based on the Intel
Core microarchitecture support non-architectural performance monitoring events.
Architectural performance events can be collected using general-purpose performance counters. Non-architectural performance events can be collected using
general-purpose performance counters (coupled with two IA32_PERFEVTSELx MSRs
for detailed event configurations), or fixed-function performance counters (see
Section 30.4.1). IA32_PERFEVTSELx MSRs are architectural; their layout is shown in
Figure 30-1. Starting with Intel Core 2 processor T 7700, fixed-function performance
counters and associated counter control and status MSR becomes part of architectural performance monitoring version 2 facilities (see also Section 30.2.2).
Non-architectural performance events in processors based on Intel Core microarchitecture use event select values that are model-specific. Valid event mask (Umask)
bits are listed in Appendix A. The UMASK field may contain sub-fields identical to
those listed in Table 30-2, Table 30-3, Table 30-4, and Table 30-5. One or more of
these sub-fields may apply to specific events on an event-by-event basis. Details are
listed in Table A-10 in Appendix A, “Performance-Monitoring Events.”
In addition, the UMASK filed may also contain a sub-field that allows detection specificity related to snoop responses. Bits of the snoop response qualification sub-field
are defined in Table 30-6.
Table 30-6. Bus Snoop Qualification Definitions within a Non-Architectural Umask
IA32_PERFEVTSELx MSRs
Bit Position 11:8
Description
Bit 11
HITM response
Bit 10
Reserved
Bit 9
HIT response
Bit 8
CLEAN response
30-18 Vol. 3B
PERFORMANCE MONITORING
There are also non-architectural events that support qualification of different types of
snoop operation. The corresponding bit field for snoop type qualification are listed in
Table 30-7.
Table 30-7. Snoop Type Qualification Definitions within a Non-Architectural Umask
IA32_PERFEVTSELx MSRs
Bit Position 9:8
Description
Bit 9
CMP2I snoops
Bit 8
CMP2S snoops
No more than one sub-field of MESI, snoop response, and snoop type qualification
sub-fields can be supported in a performance event.
NOTE
Software must write known values to the performance counters prior
to enabling the counters. The content of general-purpose counters
and fixed-function counters are undefined after INIT or RESET.
30.4.1
Fixed-function Performance Counters
Processors based on Intel Core microarchitecture provide three fixed-function performance counters. Bits beyond the width of the fixed counter are reserved and must be
written as zeros. Model-specific fixed-function performance counters on processors
that support Architectural Perfmon version 1 are 40 bits wide.
Each of the fixed-function counter is dedicated to count a pre-defined performance
monitoring events. The performance monitoring events associated with fixed-function counters and the addresses of these counters are listed in Table 30-8.
Table 30-8. Association of Fixed-Function Performance Counters with
Architectural Performance Events
Event Name
Fixed-Function PMC
PMC Address
INST_RETIRED.ANY
MSR_PERF_FIXED_CTR0/I 309H
A32_FIXED_CTR0
CPU_CLK_UNHALTED.CORE
MSR_PERF_FIXED_CTR1// 30AH
IA32_FIXED_CTR1
CPU_CLK_UNHALTED.REF
MSR_PERF_FIXED_CTR2// 30BH
IA32_FIXED_CTR2
Programming the fixed-function performance counters does not involve any of the
Vol. 3B 30-19
PERFORMANCE MONITORING
IA32_PERFEVTSELx MSRs, and does not require specifying any event masks.
Instead, the MSR MSR_PERF_FIXED_CTR_CTRL provides multiple sets of 4-bit fields;
each 4-bit field controls the operation of a fixed-function performance counter (PMC).
See Figures 30-9. Two sub-fields are defined for each control. See Figure 30-9; bit
fields are:
•
Enable field (low 2 bits in each 4-bit control) — When bit 0 is set,
performance counting is enabled in the corresponding fixed-function
performance counter to increment when the target condition associated with the
architecture performance event occurs at ring 0.
When bit 1 is set, performance counting is enabled in the corresponding fixedfunction performance counter to increment when the target condition associated
with the architecture performance event occurs at ring greater than 0.
Writing 0 to both bits stops the performance counter. Writing 11B causes the
counter to increment irrespective of privilege levels.
63
12 11
P
M
I
9 8 7
E
N
P
M
I
5 43 2 1 0
E
N
P
M
I
E
N
Cntr2 — Controls for MSR_PERF_FIXED_CTR2
Cntr1 — Controls for MSR_PERF_FIXED_CTR1
PMI — Enable PMI on overflow
Cntr0 — Controls for MSR_PERF_FIXED_CTR0
ENABLE — 0: disable; 1: OS; 2: User; 3: All ring levels
Reserved
Figure 30-9. Layout of MSR_PERF_FIXED_CTR_CTRL MSR
•
PMI field (fourth bit in each 4-bit control) — When set, the logical processor
generates an exception through its local APIC on overflow condition of the
respective fixed-function counter.
30.4.2
Global Counter Control Facilities
Processors based on Intel Core microarchitecture provides simplified performance
counter control that simplifies the most frequent operations in programming performance events, i.e. enabling/disabling event counting and checking the status of
counter overflows. This is done by the following three MSRs:
•
MSR_PERF_GLOBAL_CTRL enables/disables event counting for all or any
combination of fixed-function PMCs (MSR_PERF_FIXED_CTRx) or generalpurpose PMCs via a single WRMSR.
30-20 Vol. 3B
PERFORMANCE MONITORING
•
MSR_PERF_GLOBAL_STATUS allows software to query counter overflow
conditions on any combination of fixed-function PMCs (MSR_PERF_FIXED_CTRx)
or general-purpose PMCs via a single RDMSR.
•
MSR_PERF_GLOBAL_OVF_CTRL allows software to clear counter overflow
conditions on any combination of fixed-function PMCs (MSR_PERF_FIXED_CTRx)
or general-purpose PMCs via a single WRMSR.
MSR_PERF_GLOBAL_CTRL MSR provides single-bit controls to enable counting in
each performance counter (see Figure 30-10). Each enable bit in
MSR_PERF_GLOBAL_CTRL is AND’ed with the enable bits for all privilege levels in the
respective IA32_PERFEVTSELx or MSR_PERF_FIXED_CTR_CTRL MSRs to start/stop
the counting of respective counters. Counting is enabled if the AND’ed results is true;
counting is disabled when the result is false.
63
35 34 33 32 31
2 1 0
FIXED_CTR2 enable
FIXED_CTR1 enable
FIXED_CTR0 enable
PMC1 enable
PMC0 enable
Reserved
Figure 30-10. Layout of MSR_PERF_GLOBAL_CTRL MSR
MSR_PERF_GLOBAL_STATUS MSR provides single-bit status used by software to
query the overflow condition of each performance counter. The MSR also provides
additional status bit to indicate overflow conditions when counters are programmed
for precise-event-based sampling (PEBS). The MSR_PERF_GLOBAL_STATUS MSR
also provides a ‘sticky bit’ to indicate changes to the state of performance monitoring
hardware (see Figure 30-11). A value of 1 in bits 34:32, 1, 0 indicates an overflow
condition has occurred in the associated counter.
Vol. 3B 30-21
PERFORMANCE MONITORING
63 62
35 34 33 32 31
2 1 0
CondChgd
OvfBuffer
FIXED_CTR2 Overflow
FIXED_CTR1 Overflow
FIXED_CTR0 Overflow
PMC1 Overflow
PMC0 Overflow
Reserved
Figure 30-11. Layout of MSR_PERF_GLOBAL_STATUS MSR
When a performance counter is configured for PEBS, an overflow condition in the
counter generates a performance-monitoring interrupt this signals a PEBS event. On
a PEBS event, the processor stores data records in the buffer area (see Section
16.4.9), clears the counter overflow status, and sets the OvfBuffer bit in
MSR_PERF_GLOBAL_STATUS.
MSR_PERF_GLOBAL_OVF_CTL MSR allows software to clear overflow the indicators
for general-purpose or fixed-function counters via a single WRMSR (see
Figure 30-12). Clear overflow indications when:
•
Setting up new values in the event select and/or UMASK field for counting or
sampling
•
•
Reloading counter values to continue sampling
Disabling event counting or sampling
63 62
35 34 33 32 31
2 1 0
ClrCondChgd
ClrOvfBuffer
FIXED_CTR2 ClrOverflow
FIXED_CTR1 ClrOverflow
FIXED_CTR0 ClrOverflow
PMC1 ClrOverflow
PMC0 ClrOverflow
Reserved
Figure 30-12. Layout of MSR_PERF_GLOBAL_OVF_CTRL MSR
30-22 Vol. 3B
PERFORMANCE MONITORING
30.4.3
At-Retirement Events
Many non-architectural performance events are impacted by the speculative nature
of out-of-order execution. A subset of non-architectural performance events on
processors based on Intel Core microarchitecture are enhanced with a tagging mechanism (similar to that found in Intel NetBurst® microarchitecture) that exclude
contributions that arise from speculative execution. The at-retirement events available in processors based on Intel Core microarchitecture does not require special
MSR programming control (see Section 30.9.6, “At-Retirement Counting”), but is
limited to IA32_PMC0. See Table 30-9 for a list of events available to processors
based on Intel Core microarchitecture.
Table 30-9. At-Retirement Performance Events for Intel Core Microarchitecture
Event Name
UMask
Event Select
ITLB_MISS_RETIRED
00H
C9H
MEM_LOAD_RETIRED.L1D_MISS
01H
CBH
MEM_LOAD_RETIRED.L1D_LINE_MISS
02H
CBH
MEM_LOAD_RETIRED.L2_MISS
04H
CBH
MEM_LOAD_RETIRED.L2_LINE_MISS
08H
CBH
MEM_LOAD_RETIRED.DTLB_MISS
10H
CBH
30.4.4
Precise Event Based Sampling (PEBS)
Processors based on Intel Core microarchitecture also support precise event based
sampling (PEBS). This feature was introduced by processors based on Intel NetBurst
microarchitecture.
PEBS uses a debug store mechanism and a performance monitoring interrupt to
store a set of architectural state information for the processor. The information
provides architectural state of the instruction executed after the instruction that
caused the event (See Section 30.4.4.2).
In cases where the same instruction causes BTS and PEBS to be activated, PEBS is
processed before BTS are processed. The PMI request is held until the processor
completes processing of PEBS and BTS.
For processors based on Intel Core microarchitecture, events that support precise
sampling are listed in Table 30-10. The procedure for detecting availability of PEBS is
the same as described in Section 30.9.7.1.
Table 30-10. PEBS Performance Events for Intel Core Microarchitecture
Event Name
UMask
Event Select
INSTR_RETIRED.ANY_P
00H
C0H
X87_OPS_RETIRED.ANY
FEH
C1H
Vol. 3B 30-23
PERFORMANCE MONITORING
Table 30-10. PEBS Performance Events for Intel Core Microarchitecture (Contd.)
Event Name
UMask
Event Select
BR_INST_RETIRED.MISPRED
00H
C5H
SIMD_INST_RETIRED.ANY
1FH
C7H
MEM_LOAD_RETIRED.L1D_MISS
01H
CBH
MEM_LOAD_RETIRED.L1D_LINE_MISS
02H
CBH
MEM_LOAD_RETIRED.L2_MISS
04H
CBH
MEM_LOAD_RETIRED.L2_LINE_MISS
08H
CBH
MEM_LOAD_RETIRED.DTLB_MISS
10H
CBH
30.4.4.1
Setting up the PEBS Buffer
For processors based on Intel Core microarchitecture, PEBS is available using
IA32_PMC0 only. Use the following procedure to set up the processor and
IA32_PMC0 counter for PEBS:
1. Set up the precise event buffering facilities. Place values in the precise event
buffer base, precise event index, precise event absolute maximum, precise event
interrupt threshold, and precise event counter reset fields of the DS buffer
management area. In processors based on Intel Core microarchitecture, PEBS
records consist of 64-bit address entries. See Figure 16-8 to set up the precise
event records buffer in memory.
2. Enable PEBS. Set the Enable PEBS on PMC0 flag (bit 0) in IA32_PEBS_ENABLE
MSR.
3. Set up the IA32_PMC0 performance counter and IA32_PERFEVTSEL0 for an
event listed in Table 30-10.
30.4.4.2
PEBS Record Format
The PEBS record format may be extended across different processor implementations. The IA32_PERF_CAPABILITES MSR defines a mechanism for software to
handle the evolution of PEBS record format in processors that support architectural
performance monitoring with version id equals 2 or higher. The bit fields of
IA32_PERF_CAPABILITES are defined in Table B-2 of Appendix B, “Model-Specific
Registers (MSRs)”. The relevant bit fields that governs PEBS are:
•
PEBSTrap [bit 6]: When set, PEBS recording is trap-like. After the PEBS-enabled
counter has overflowed, PEBS record is recorded for the next PEBS-able event at
the completion of the sampled instruction causing the PEBS event. When clear,
PEBS recording is fault-like. The PEBS record is recorded before the sampled
instruction causing the PEBS event.
•
PEBSSaveArchRegs [bit 7]: When set, PEBS will save architectural register and
state information according to the encoded value of the PEBSRecordFormat field.
On processors based on Intel Core microarchitecture, this bit is always 1
30-24 Vol. 3B
PERFORMANCE MONITORING
•
PEBSRecordFormat [bits 11:8]: Valid encodings are:
— 0000B: Only general-purpose registers, instruction pointer and RFLAGS
registers are saved in each PEBS record (seeSection 30.9.7).
30.4.4.3
Writing a PEBS Interrupt Service Routine
The PEBS facilities share the same interrupt vector and interrupt service routine
(called the DS ISR) with the non-precise event-based sampling and BTS facilities. To
handle PEBS interrupts, PEBS handler code must be included in the DS ISR. See
Section 16.4.9.1, “DS Save Area and IA-32e Mode Operation,” for guidelines when
writing the DS ISR.
The service routine can query MSR_PERF_GLOBAL_STATUS to determine which
counter(s) caused of overflow condition. The service routine should clear overflow
indicator by writing to MSR_PERF_GLOBAL_OVF_CTL.
A comparison of the sequence of requirements to program PEBS for processors based
on Intel Core and Intel NetBurst microarchitectures is listed in Table 30-11.
Table 30-11. Requirements to Program PEBS
For Processors based on Intel
Core microarchitecture
For Processors based on Intel
NetBurst microarchitecture
Verify PEBS support of
processor/OS
• IA32_MISC_ENABLE.EMON_AVAILABE (bit 7) is set.
• IA32_MISC_ENABLE.PEBS_UNAVAILABE (bit 12) is clear.
Ensure counters are in
disabled
On initial set up or changing event
configurations, write
MSR_PERF_GLOBAL_CTRL MSR
(0x38F) with 0.
Optional
On subsequent entries:
• Clear all counters if “Counter
Freeze on PMI“ is not enabled.
• If IA32_DebugCTL.Freeze is
enabled, counters are
automatically disabled.
Counters MUST be stopped before
writing.1
Disable PEBS.
Clear ENABLE PMC0 bit in
IA32_PEBS_ENABLE MSR
(0x3F1).
Optional
Check overflow
conditions.
Check
MSR_PERF_GLOBAL_STATUS MSR
(0x 38E) handle any overflow
conditions.
Check OVF flag of each CCCR for
overflow condition
Vol. 3B 30-25
PERFORMANCE MONITORING
Table 30-11. Requirements to Program PEBS (Contd.)
For Processors based on Intel
Core microarchitecture
For Processors based on Intel
NetBurst microarchitecture
Clear overflow status.
Clear
MSR_PERF_GLOBAL_STATUS MSR
(0x 38E) using
IA32_PERF_GLOBAL_OVF_CTRL
MSR (0x390).
Clear OVF flag of each CCCR.
Write “sample-after“
values.
Configure the counter(s) with the sample after value.
Configure specific counter
configuration MSR.
• Set local enable bit 22 - 1.
• Do NOT set local counter
PMI/INT bit, bit 20 - 0.
• Event programmed must be
PEBS capable.
Allocate buffer for PEBS
states.
Allocate a buffer in memory for the precise information.
Program the
IA32_DS_AREA MSR.
Program the IA32_DS_AREA MSR.
Configure the PEBS buffer
management records.
Configure the PEBS buffer management records in the DS buffer
management area.
Configure/Enable PEBS.
Set Enable PMC0 bit in
IA32_PEBS_ENABLE MSR
(0x3F1).
Configure MSR_PEBS_ENABLE,
MSR_PEBS_MATRIX_VERT and
MSR_PEBS_MATRIX_HORZ as
needed.
Enable counters.
Set Enable bits in
MSR_PERF_GLOBAL_CTRL MSR
(0x38F).
Set each CCCR enable bit 12 - 1.
• Set appropriate OVF_PMI bits 1.
• Only CCCR for
MSR_IQ_COUNTER4 support
PEBS.
NOTES:
1. Counters read while enabled are not guaranteed to be precise with event counts that occur in timing proximity to the RDMSR.
30.4.4.4
Re-configuring PEBS Facilities
When software needs to reconfigure PEBS facilities, it should allow a quiescent period
between stopping the prior event counting and setting up a new PEBS event. The
quiescent period is to allow any latent residual PEBS records to complete its capture
at their previously specified buffer address (provided by IA32_DS_AREA).
30-26 Vol. 3B
PERFORMANCE MONITORING
30.5
PERFORMANCE MONITORING (PROCESSORS BASED
ON INTEL® ATOM™ MICROARCHITECTURE)
Intel Atom processor family supports architectural performance monitoring capability with version ID 3 (see Section 30.2.2.2) and a host of non-architectural monitoring capabilities. The initial implementation of Intel Atom processor family provides
two general-purpose performance counters (IA32_PMC0, IA32_PMC1) and three
fixed-function performance counters (IA32_FIXED_CTR0, IA32_FIXED_CTR1,
IA32_FIXED_CTR2).
Non-architectural performance monitoring in Intel Atom processor family uses the
IA32_PERFEVTSELx MSR to configure a set of non-architecture performance monitoring events to be counted by the corresponding general-purpose performance
counter. The list of non-architectural performance monitoring events is listed in Table
A-11.
Architectural and non-architectural performance monitoring events in Intel Atom
processor family support thread qualification using bit 21 of IA32_PERFEVTSELx
MSR.
The bit fields within each IA32_PERFEVTSELx MSR are defined in Figure 30-6 and
described in Section 30.2.1.1 and Section 30.2.2.2.
Valid event mask (Umask) bits are listed in Appendix A. The UMASK field may contain
sub-fields that provide the same qualifying actions like those listed in Table 30-2,
Table 30-3, Table 30-4, and Table 30-5. One or more of these sub-fields may apply to
specific events on an event-by-event basis. Details are listed in Table A-11 in
Appendix A, “Performance-Monitoring Events.” Precise Event Based Monitoring is
supported using IA32_PMC0 (see also Section 16.4.9, “BTS and DS Save Area”).
30.6
PERFORMANCE MONITORING FOR PROCESSORS
BASED ON INTEL® MICROARCHITECTURE CODE
NAME NEHALEM
Intel Core i7 processor family1 supports architectural performance monitoring capability with version ID 3 (see Section 30.2.2.2) and a host of non-architectural monitoring capabilities. The Intel Core i7 processor family is based on Intel®
microarchitecture code name Nehalem, and provides four general-purpose performance counters (IA32_PMC0, IA32_PMC1, IA32_PMC2, IA32_PMC3) and three
fixed-function performance counters (IA32_FIXED_CTR0, IA32_FIXED_CTR1,
IA32_FIXED_CTR2) in the processor core.
1. Intel Xeon processor 5500 series and 3400 series are also based on Intel microarchitecture code
name Nehalem, so the performance monitoring facilities described in this section generally also
apply.
Vol. 3B 30-27
PERFORMANCE MONITORING
Non-architectural performance monitoring in Intel Core i7 processor family uses the
IA32_PERFEVTSELx MSR to configure a set of non-architecture performance monitoring events to be counted by the corresponding general-purpose performance
counter. The list of non-architectural performance monitoring events is listed in Table
A-11. Non-architectural performance monitoring events fall into two broad categories:
•
Performance monitoring events in the processor core: These include many
events that are similar to performance monitoring events available to processor
based on Intel Core microarchitecture. Additionally, there are several enhancements in the performance monitoring capability for detecting microarchitectural
conditions in the processor core or in the interaction of the processor core to the
off-core sub-systems in the physical processor package. The off-core subsystems in the physical processor package is loosely referred to as “uncore“.
•
Performance monitoring events in the uncore: The uncore sub-system is shared
by more than one processor cores in the physical processor package. It provides
additional performance monitoring facility outside of IA32_PMCx and
performance monitoring events that are specific to the uncore sub-system.
Architectural and non-architectural performance monitoring events in Intel Core i7
processor family support thread qualification using bit 21 of IA32_PERFEVTSELx
MSR.
The bit fields within each IA32_PERFEVTSELx MSR are defined in Figure 30-6 and
described in Section 30.2.1.1 and Section 30.2.2.2.
63 62 61 60
353433 3231
8 7 6 5 43 2 1 0
CHG (R/W)
OVF_PMI (R/W)
OVF_FC2 (R/O)
OVF_FC1 (R/O)
OVF_FC0 (R/O)
OVF_PC7 (R/O), if CCNT>7
OVF_PC6 (R/O), if CCNT>6
OVF_PC5 (R/O), if CCNT>5
OVF_PC4 (R/O), if CCNT>4
OVF_PC3 (R/O)
OVF_PC2 (R/O)
OVF_PC1 (R/O)
OVF_PC0 (R/O)
Reserved
RESET Value — 0x00000000_00000000
CCNT: CPUID.AH:EAX[15:8]
Figure 30-13. IA32_PERF_GLOBAL_STATUS MSR
30-28 Vol. 3B
PERFORMANCE MONITORING
30.6.1
Enhancements of Performance Monitoring in the Processor
Core
The notable enhancements in the monitoring of performance events in the processor
core include:
•
Four general purpose performance counters, IA32_PMCx, associated counter
configuration MSRs, IA32_PERFEVTSELx, and global counter control MSR
supporting simplified control of four counters. Each of the four performance
counter can support precise event based sampling (PEBS) and thread-qualification of architectural and non-architectural performance events. Width of
IA32_PMCx supported by hardware has been increased. The width of counter
reported by CPUID.0AH:EAX[23:16] is 48 bits. The PEBS facility in Intel microarchitecture code name Nehalem has been enhanced to include new data format to
capture additional information, such as load latency.
•
Load latency sampling facility. Average latency of memory load operation can be
sampled using load-latency facility in processors based on Intel microarchitecture code name Nehalem. The facility can measure average latency of load
micro-operations from dispatch to when data is globally observable (GO). This
facility is used in conjunction with the PEBS facility.
•
Off-core response counting facility. This facility in the processor core allows
software to count certain transaction responses between the processor core to
sub-systems outside the processor core (uncore). Counting off-core response
requires additional event qualification configuration facility in conjunction with
IA32_PERFEVTSELx. Two off-core response MSRs are provided to use in
conjunction with specific event codes that must be specified with
IA32_PERFEVTSELx.
30.6.1.1
Precise Event Based Sampling (PEBS)
All four general-purpose performance counters, IA32_PMCx, can be used for PEBS if
the performance event supports PEBS. Software uses IA32_MISC_ENABLE[7] and
IA32_MISC_ENABLE[12] to detect whether the performance monitoring facility and
PEBS functionality are supported in the processor. The MSR IA32_PEBS_ENABLE
provides 4 bits that software must use to enable which IA32_PMCx overflow condition will cause the PEBS record to be captured.
Additionally, the PEBS record is expanded to allow latency information to be
captured. The MSR IA32_PEBS_ENABLE provides 4 additional bits that software must
use to enable latency data recording in the PEBS record upon the respective
IA32_PMCx overflow condition. The layout of IA32_PEBS_ENABLE for processors
based on Intel microarchitecture code name Nehalem is shown in Figure 30-14.
When a counter is enabled to capture machine state (PEBS_EN_PMCx = 1), the
processor will write machine state information to a memory buffer specified by software as detailed below. When the counter IA32_PMCx overflows from maximum
count to zero, the PEBS hardware is armed.
Vol. 3B 30-29
PERFORMANCE MONITORING
36 3534 33 32 31
63
8 7 6 5 43 2 1 0
LL_EN_PMC3 (R/W)
LL_EN_PMC2 (R/W)
LL_EN_PMC1 (R/W)
LL_EN_PMC0 (R/W)
PEBS_EN_PMC3 (R/W)
PEBS_EN_PMC2 (R/W)
PEBS_EN_PMC1 (R/W)
PEBS_EN_PMC0 (R/W)
Reserved
RESET Value — 0x00000000_00000000
Figure 30-14. Layout of IA32_PEBS_ENABLE MSR
Upon occurrence of the next PEBS event, the PEBS hardware triggers an assist and
causes a PEBS record to be written. The format of the PEBS record is indicated by the
bit field IA32_PERF_CAPABILITIES[11:8] (see Figure 30-39).
The behavior of PEBS assists is reported by IA32_PERF_CAPABILITIES[6] (see
Figure 30-39). The return instruction pointer (RIP) reported in the PEBS record will
point to the instruction after (+1) the instruction that causes the PEBS assist. The
machine state reported in the PEBS record is the machine state after the instruction
that causes the PEBS assist is retired. For instance, if the instructions:
mov eax, [eax] ; causes PEBS assist
nop
are executed, the PEBS record will report the address of the nop, and the value of
EAX in the PEBS record will show the value read from memory, not the target address
of the read operation.
The PEBS record format is shown in Table 30-12, and each field in the PEBS record is
64 bits long. The PEBS record format, along with debug/store area storage format,
does not change regardless of IA-32e mode is active or not.
CPUID.01H:ECX.DTES64[bit 2] reports the processor’s support for 64-bit
debug/store area storage format is invariant to IA-32e mode.
Table 30-12. PEBS Record Format for Intel Core i7 Processor Family
Byte Offset
Field
Byte Offset
Field
0x0
R/EFLAGS
0x58
R9
30-30 Vol. 3B
PERFORMANCE MONITORING
Table 30-12. PEBS Record Format for Intel Core i7 Processor Family
Byte Offset
Field
Byte Offset
Field
0x8
R/EIP
0x60
R10
0x10
R/EAX
0x68
R11
0x18
R/EBX
0x70
R12
0x20
R/ECX
0x78
R13
0x28
R/EDX
0x80
R14
0x30
R/ESI
0x88
R15
0x38
R/EDI
0x90
IA32_PERF_GLOBAL_STATUS
0x40
R/EBP
0x98
Data Linear Address
0x48
R/ESP
0xA0
Data Source Encoding
0x50
R8
0xA8
Latency value (core cycles)
In IA-32e mode, the full 64-bit value is written to the register. If the processor is not
operating in IA-32e mode, 32-bit value is written to registers with bits 63:32 zeroed.
Registers not defined when the processor is not in IA-32e mode are written to zero.
Bytes 0xAF:0x90 are enhancement to the PEBS record format. Support for this
enhanced PEBS record format is indicated by IA32_PERF_CAPABILITIES[11:8]
encoding of 0001B.
The value written to bytes 0x97:0x90 is the state of the
IA32_PERF_GLOBAL_STATUS register before the PEBS assist occurred. This value is
written so software can determine which counters overflowed when this PEBS record
was written. Note that this field indicates the overflow status for all counters, regardless of whether they were programmed for PEBS or not.
Programming PEBS Facility
Only a subset of non-architectural performance events in the processor support
PEBS. The subset of precise events are listed in Table 30-10. In addition to using
IA32_PERFEVTSELx to specify event unit/mask settings and setting the EN_PMCx bit
in the IA32_PEBS_ENABLE register for the respective counter, the software must also
initialize the DS_BUFFER_MANAGEMENT_AREA data structure in memory to support
capturing PEBS records for precise events.
NOTE
PEBS events are only valid when the following fields of
IA32_PERFEVTSELx are all zero: AnyThread, Edge, Invert, CMask.
The beginning linear address of the DS_BUFFER_MANAGEMENT_AREA data structure
must be programmed into the IA32_DS_AREA register. The layout of the
DS_BUFFER_MANAGEMENT_AREA is shown in Figure 30-15.
Vol. 3B 30-31
PERFORMANCE MONITORING
•
PEBS Buffer Base: This field is programmed with the linear address of the first
byte of the PEBS buffer allocated by software. The processor reads this field to
determine the base address of the PEBS buffer. Software should allocate this
memory from the non-paged pool.
IA32_DS_AREA MSR
DS Buffer Management Area
BTS Buffer Base
0H
BTS Index
8H
BTS Absolute
Maximum
BTS Interrupt
Threshold
BTS Buffer
Branch Record 0
10H
Branch Record 1
18H
PEBS Buffer Base 20H
PEBS Index
PEBS Absolute
Maximum
PEBS Interrupt
Threshold
PEBS
Counter0 Reset
PEBS
Counter1 Reset
PEBS
Counter2 Reset
PEBS
Counter3 Reset
Reserved
28H
30H
38H
Branch Record n
40H
48H
PEBS Buffer
PEBS Record 0
50H
58H
PEBS Record 1
60H
PEBS Record n
Figure 30-15. PEBS Programming Environment
•
PEBS Index: This field is initially programmed with the same value as the PEBS
Buffer Base field, or the beginning linear address of the PEBS buffer. The
processor reads this field to determine the location of the next PEBS record to
write to. After a PEBS record has been written, the processor also updates this
field with the address of the next PEBS record to be written. The figure above
illustrates the state of PEBS Index after the first PEBS record is written.
30-32 Vol. 3B
PERFORMANCE MONITORING
•
PEBS Absolute Maximum: This field represents the absolute address of the
maximum length of the allocated PEBS buffer plus the starting address of the
PEBS buffer. The processor will not write any PEBS record beyond the end of
PEBS buffer, when PEBS Index equals PEBS Absolute Maximum. No signaling
is generated when PEBS buffer is full. Software must reset the PEBS Index field
to the beginning of the PEBS buffer address to continue capturing PEBS records.
•
PEBS Interrupt Threshold: This field specifies the threshold value to trigger a
performance interrupt and notify software that the PEBS buffer is nearly full. This
field is programmed with the linear address of the first byte of the PEBS record
within the PEBS buffer that represents the threshold record. After the processor
writes a PEBS record and updates PEBS Index, if the PEBS Index reaches the
threshold value of this field, the processor will generate a performance interrupt.
This is the same interrupt that is generated by a performance counter overflow,
as programmed in the Performance Monitoring Counters vector in the Local
Vector Table of the Local APIC. When a performance interrupt due to PEBS buffer
full is generated, the IA32_PERF_GLOBAL_STATUS.PEBS_Ovf bit will be set.
•
PEBS CounterX Reset: This field allows software to set up PEBS counter
overflow condition to occur at a rate useful for profiling workload, thereby
generating multiple PEBS records to facilitate characterizing the profile the
execution of test code. After each PEBS record is written, the processor checks
each counter to see if it overflowed and was enabled for PEBS (the corresponding
bit in IA32_PEBS_ENABLED was set). If these conditions are met, then the reset
value for each overflowed counter is loaded from the DS Buffer Management
Area. For example, if counter IA32_PMC0 caused a PEBS record to be written,
then the value of “PEBS Counter 0 Reset” would be written to counter
IA32_PMC0. If a counter is not enabled for PEBS, its value will not be modified by
the PEBS assist.
Performance Counter Prioritization
Performance monitoring interrupts are triggered by a counter transitioning from
maximum count to zero (assuming IA32_PerfEvtSelX.INT is set). This same transition will cause PEBS hardware to arm, but not trigger. PEBS hardware triggers upon
detection of the first PEBS event after the PEBS hardware has been armed (a 0 to 1
transition of the counter). At this point, a PEBS assist will be undertaken by the
processor.
Performance counters (fixed and general-purpose) are prioritized in index order. That
is, counter IA32_PMC0 takes precedence over all other counters. Counter
IA32_PMC1 takes precedence over counters IA32_PMC2 and IA32_PMC3, and so on.
This means that if simultaneous overflows or PEBS assists occur, the appropriate
action will be taken for the highest priority performance counter. For example, if
IA32_PMC1 cause an overflow interrupt and IA32_PMC2 causes an PEBS assist
simultaneously, then the overflow interrupt will be serviced first.
The PEBS threshold interrupt is triggered by the PEBS assist, and is by definition
prioritized lower than the PEBS assist. Hardware will not generate separate interrupts
for each counter that simultaneously overflows. General-purpose performance
counters are prioritized over fixed counters.
Vol. 3B 30-33
PERFORMANCE MONITORING
If a counter is programmed with a precise (PEBS-enabled) event and programmed to
generate a counter overflow interrupt, the PEBS assist is serviced before the counter
overflow interrupt is serviced. If in addition the PEBS interrupt threshold is met, the
threshold interrupt is generated after the PEBS assist completes, followed by the
counter overflow interrupt (two separate interrupts are generated).
Uncore counters may be programmed to interrupt one or more processor cores (see
Section 30.6.2). It is possible for interrupts posted from the uncore facility to occur
coincident with counter overflow interrupts from the processor core. Software must
check core and uncore status registers to determine the exact origin of counter overflow interrupts.
30.6.1.2
Load Latency Performance Monitoring Facility
The load latency facility provides software a means to characterize the average load
latency to different levels of cache/memory hierarchy. This facility requires processor
supporting enhanced PEBS record format in the PEBS buffer, see Table 30-12. The
facility measures latency from micro-operation (uop) dispatch to when data is
globally observable (GO).
To use this feature software must assure:
•
One of the IA32_PERFEVTSELx MSR is programmed to specify the event unit
MEM_INST_RETIRED, and the LATENCY_ABOVE_THRESHOLD event mask must
be specified (IA32_PerfEvtSelX[15:0] = 0x100H). The corresponding counter
IA32_PMCx will accumulate event counts for architecturally visible loads which
exceed the programmed latency threshold specified separately in a MSR. Stores
are ignored when this event is programmed. The CMASK or INV fields of the
IA32_PerfEvtSelX register used for counting load latency must be 0. Writing
other values will result in undefined behavior.
•
The MSR_PEBS_LD_LAT_THRESHOLD MSR is programmed with the desired
latency threshold in core clock cycles. Loads with latencies greater than this
value are eligible for counting and latency data reporting. The minimum value
that may be programmed in this register is 3 (the minimum detectable load
latency is 4 core clock cycles).
•
The PEBS enable bit in the IA32_PEBS_ENABLE register is set for the corresponding IA32_PMCx counter register. This means that both the PEBS_EN_CTRX
and LL_EN_CTRX bits must be set for the counter(s) of interest. For example, to
enable load latency on counter IA32_PMC0, the IA32_PEBS_ENABLE register
must be programmed with the 64-bit value 0x00000001.00000001.
When the load-latency facility is enabled, load operations are randomly selected by
hardware and tagged to carry information related to data source locality and latency.
Latency and data source information of tagged loads are updated internally.
When a PEBS assist occurs, the last update of latency and data source information
are captured by the assist and written as part of the PEBS record. The PEBS sample
after value (SAV), specified in PEBS CounterX Reset, operates orthogonally to the
tagging mechanism. Loads are randomly tagged to collect latency data. The SAV
30-34 Vol. 3B
PERFORMANCE MONITORING
controls the number of tagged loads with latency information that will be written into
the PEBS record field by the PEBS assists. The load latency data written to the PEBS
record will be for the last tagged load operation which retired just before the PEBS
assist was invoked.
The load-latency information written into a PEBS record (see Table 30-12, bytes
AFH:98H) consists of:
•
Data Linear Address: This is the linear address of the target of the load
operation.
•
Latency Value: This is the elapsed cycles of the tagged load operation between
dispatch to GO, measured in processor core clock domain.
•
Data Source : The encoded value indicates the origin of the data obtained by the
load instruction. The encoding is shown in Table 30-13. In the descriptions local
memory refers to system memory physically attached to a processor package,
and remote memory referrals to system memory physically attached to another
processor package.
Table 30-13. Data Source Encoding for Load Latency Record
Encoding
Description
0x0
Unknown L3 cache miss
0x1
Minimal latency core cache hit. This request was satisfied by the L1 data cache.
0x2
Pending core cache HIT. Outstanding core cache miss to same cache-line address
was already underway.
0x3
This data request was satisfied by the L2.
0x4
L3 HIT. Local or Remote home requests that hit L3 cache in the uncore with no
coherency actions required (snooping).
0x5
L3 HIT. Local or Remote home requests that hit the L3 cache and was serviced by
another processor core with a cross core snoop where no modified copies were
found. (clean).
0x6
L3 HIT. Local or Remote home requests that hit the L3 cache and was serviced by
another processor core with a cross core snoop where modified copies were found.
(HITM).
0x7
Reserved
0x8
L3 MISS. Local homed requests that missed the L3 cache and was serviced by
forwarded data following a cross package snoop where no modified copies found.
(Remote home requests are not counted).
0x9
Reserved
0xA
L3 MISS. Local home requests that missed the L3 cache and was serviced by local
DRAM (go to shared state).
Vol. 3B 30-35
PERFORMANCE MONITORING
Table 30-13. Data Source Encoding for Load Latency Record (Contd.)
Encoding
Description
0xB
L3 MISS. Remote home requests that missed the L3 cache and was serviced by
remote DRAM (go to shared state).
0xC
L3 MISS. Local home requests that missed the L3 cache and was serviced by local
DRAM (go to exclusive state).
0xD
L3 MISS. Remote home requests that missed the L3 cache and was serviced by
remote DRAM (go to exclusive state).
0xE
I/O, Request of input/output operation
0xF
The request was to un-cacheable memory.
The layout of MSR_PEBS_LD_LAT_THRESHOLD is shown in Figure 30-16.
63
1615
0
THRHLD - Load latency threshold
Reserved
RESET Value — 0x00000000_00000000
Figure 30-16. Layout of MSR_PEBS_LD_LAT MSR
Bits 15:0 specifies the threshold load latency in core clock cycles. Performance
events with latencies greater than this value are counted in IA32_PMCx and their
latency information is reported in the PEBS record. Otherwise, they are ignored. The
minimum value that may be programmed in this field is 3.
30.6.1.3
Off-core Response Performance Monitoring in the Processor Core
Performance an event using off-core response facility can program any of the four
IA32_PERFEVTSELx MSR with specific event codes and predefine mask bit value.
Each event code for off-core response monitoring requires programming an associated configuration MSR, MSR_OFFCORE_RSP_0. There is only one off-core response
configuration MSR. Table 30-14 lists the event code, mask value and additional offcore configuration MSR that must be programmed to count off-core response events
using IA32_PMCx.
30-36 Vol. 3B
PERFORMANCE MONITORING
Table 30-14. Off-Core Response Event Encoding
Event code in
IA32_PERFEVTSELx
Mask Value in
IA32_PERFEVTSELx
Required Off-core Response MSR
0xB7
0x01
MSR_OFFCORE_RSP_0 (address 0x1A6)
The layout of MSR_OFFCORE_RSP_0 is shown in Figure 30-17. Bits 7:0 specifies the
request type of a transaction request to the uncore. Bits 15:8 specifies the response
of the uncore subsystem.
63
15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
RESPONSE TYPE — NON_DRAM (R/W)
RESPONSE TYPE — LOCAL_DRAM (R/W)
RESPONSE TYPE — REMOTE_DRAM (R/W)
RESPONSE TYPE — REMOTE_CACHE_FWD (R/W)
RESPONSE TYPE — RESERVED
RESPONSE TYPE — OTHER_CORE_HITM (R/W)
RESPONSE TYPE — OTHER_CORE_HIT_SNP (R/W)
RESPONSE TYPE — UNCORE_HIT (R/W)
REQUEST TYPE — OTHER (R/W)
REQUEST TYPE — PF_IFETCH (R/W)
REQUEST TYPE — PF_RFO (R/W)
REQUEST TYPE — PF_DATA_RD (R/W)
REQUEST TYPE — WB (R/W)
REQUEST TYPE — DMND_IFETCH (R/W)
REQUEST TYPE — DMND_RFO (R/W)
REQUEST TYPE — DMND_DATA_RD (R/W)
Reserved
RESET Value — 0x00000000_00000000
Figure 30-17. Layout of MSR_OFFCORE_RSP_0 and MSR_OFFCORE_RSP_1 to
Configure Off-core Response Events
Table 30-15. MSR_OFFCORE_RSP_0 and MSR_OFFCORE_RSP_1 Bit Field Definition
Bit Name
Offset
Description
DMND_DATA_RD
0
(R/W). Counts the number of demand and DCU prefetch data reads
of full and partial cachelines as well as demand data page table
entry cacheline reads. Does not count L2 data read prefetches or
instruction fetches.
Vol. 3B 30-37
PERFORMANCE MONITORING
Table 30-15. MSR_OFFCORE_RSP_0 and MSR_OFFCORE_RSP_1 Bit Field Definition
Bit Name
Offset
Description
DMND_RFO
1
(R/W). Counts the number of demand and DCU prefetch reads for
ownership (RFO) requests generated by a write to data cacheline.
Does not count L2 RFO.
DMND_IFETCH
2
(R/W). Counts the number of demand and DCU prefetch instruction
cacheline reads. Does not count L2 code read prefetches.
WB
3
(R/W). Counts the number of writeback (modified to exclusive)
transactions.
PF_DATA_RD
4
(R/W). Counts the number of data cacheline reads generated by L2
prefetchers.
PF_RFO
5
(R/W). Counts the number of RFO requests generated by L2
prefetchers.
PF_IFETCH
6
(R/W). Counts the number of code reads generated by L2
prefetchers.
OTHER
7
(R/W). Counts one of the following transaction types, including L3
invalidate, I/O, full or partial writes, WC or non-temporal stores,
CLFLUSH, Fences, lock, unlock, split lock.
UNCORE_HIT
8
(R/W). L3 Hit: local or remote home requests that hit L3 cache in the
uncore with no coherency actions required (snooping).
OTHER_CORE_HI
T_SNP
9
(R/W). L3 Hit: local or remote home requests that hit L3 cache in the
uncore and was serviced by another core with a cross core snoop
where no modified copies were found (clean).
OTHER_CORE_HI
TM
10
(R/W). L3 Hit: local or remote home requests that hit L3 cache in the
uncore and was serviced by another core with a cross core snoop
where modified copies were found (HITM).
Reserved
11
Reserved
REMOTE_CACHE_ 12
FWD
(R/W). L3 Miss: local homed requests that missed the L3 cache and
was serviced by forwarded data following a cross package snoop
where no modified copies found. (Remote home requests are not
counted)
REMOTE_DRAM
13
(R/W). L3 Miss: remote home requests that missed the L3 cache and
were serviced by remote DRAM.
LOCAL_DRAM
14
(R/W). L3 Miss: local home requests that missed the L3 cache and
were serviced by local DRAM.
NON_DRAM
15
(R/W). Non-DRAM requests that were serviced by IOH.
30-38 Vol. 3B
PERFORMANCE MONITORING
30.6.2
Performance Monitoring Facility in the Uncore
The “uncore” in Intel microarchitecture code name Nehalem refers to subsystems in
the physical processor package that are shared by multiple processor cores. Some of
the sub-systems in the uncore include the L3 cache, Intel QuickPath Interconnect link
logic, and integrated memory controller. The performance monitoring facilities inside
the uncore operates in the same clock domain as the uncore (U-clock domain), which
is usually different from the processor core clock domain. The uncore performance
monitoring facilities described in this section apply to Intel Xeon processor 5500
series and processors with the following CPUID signatures: 06_1AH, 06_1EH,
06_1FH (see Appendix B). An overview of the uncore performance monitoring facilities is described separately.
The performance monitoring facilities available in the U-clock domain consist of:
•
Eight General-purpose counters (MSR_UNCORE_PerfCntr0 through
MSR_UNCORE_PerfCntr7). The counters are 48 bits wide. Each counter is
associated with a configuration MSR, MSR_UNCORE_PerfEvtSelx, to specify
event code, event mask and other event qualification fields. A set of global
uncore performance counter enabling/overflow/status control MSRs are also
provided for software.
•
Performance monitoring in the uncore provides an address/opcode match MSR
that provides event qualification control based on address value or QPI command
opcode.
•
One fixed-function counter, MSR_UNCORE_FixedCntr0. The fixed-function
uncore counter increments at the rate of the U-clock when enabled.
The frequency of the uncore clock domain can be determined from the uncore
clock ratio which is available in the PCI configuration space register at offset C0H
under device number 0 and Function 0.
30.6.2.1
Uncore Performance Monitoring Management Facility
MSR_UNCORE_PERF_GLOBAL_CTRL provides bit fields to enable/disable generalpurpose and fixed-function counters in the uncore. Figure 30-18 shows the layout of
MSR_UNCORE_PERF_GLOBAL_CTRL for an uncore that is shared by four processor
cores in a physical package.
•
EN_PCn (bit n, n = 0, 7): When set, enables counting for the general-purpose
uncore counter MSR_UNCORE_PerfCntr n.
•
EN_FC0 (bit 32): When set, enables counting for the fixed-function uncore
counter MSR_UNCORE_FixedCntr0.
•
EN_PMI_COREn (bit n, n = 0, 3 if four cores are present): When set, processor
core n is programmed to receive an interrupt signal from any interrupt enabled
uncore counter. PMI delivery due to an uncore counter overflow is enabled by
setting IA32_DEBUG_CTL.Offcore_PMI_EN to 1.
•
PMI_FRZ (bit 63): When set, all U-clock uncore counters are disabled when any
one of them signals a performance interrupt. Software must explicitly re-enable
Vol. 3B 30-39
PERFORMANCE MONITORING
the counter by setting the enable bits in MSR_UNCORE_PERF_GLOBAL_CTRL
upon exit from the ISR.
63 62
51 50 49 48
32 31
8 7 6 5 43 2 1 0
PMI_FRZ (R/W)
EN_PMI_CORE3 (R/W)
EN_PMI_CORE2 (R/W)
EN_PMI_CORE1 (R/W)
EN_PMI_CORE0 (R/W)
EN_FC0 (R/W)
EN_PC7 (R/W)
EN_PC6 (R/W)
EN_PC5 (R/W)
EN_PC4 (R/W)
EN_PC3 (R/W)
EN_PC2 (R/W)
EN_PC1 (R/W)
EN_PC0 (R/W)
Reserved
RESET Value — 0x00000000_00000000
Figure 30-18. Layout of MSR_UNCORE_PERF_GLOBAL_CTRL MSR
MSR_UNCORE_PERF_GLOBAL_STATUS provides overflow status of the U-clock
performance counters in the uncore. This is a read-only register. If an overflow status
bit is set the corresponding counter has overflowed. The register provides a condition
change bit (bit 63) which can be quickly checked by software to determine if a significant change has occurred since the last time the condition change status was
cleared. Figure 30-19 shows the layout of MSR_UNCORE_PERF_GLOBAL_STATUS.
•
OVF_PCn (bit n, n = 0, 7): When set, indicates general-purpose uncore counter
MSR_UNCORE_PerfCntr n has overflowed.
•
OVF_FC0 (bit 32): When set, indicates the fixed-function uncore counter
MSR_UNCORE_FixedCntr0 has overflowed.
•
OVF_PMI (bit 61): When set indicates that an uncore counter overflowed and
generated an interrupt request.
•
CHG (bit 63): When set indicates that at least one status bit in
MSR_UNCORE_PERF_GLOBAL_STATUS register has changed state.
MSR_UNCORE_PERF_GLOBAL_OVF_CTRL allows software to clear the status bits in
the UNCORE_PERF_GLOBAL_STATUS register. This is a write-only register, and individual status bits in the global status register are cleared by writing a binary one to
the corresponding bit in this register. Writing zero to any bit position in this register
has no effect on the uncore PMU hardware.
30-40 Vol. 3B
PERFORMANCE MONITORING
63 62 61 60
32 31
8 7 6 5 43 2 1 0
CHG (R/W)
OVF_PMI (R/W)
OVF_FC0 (R/O)
OVF_PC7 (R/O)
OVF_PC6 (R/O)
OVF_PC5 (R/O)
OVF_PC4 (R/O)
OVF_PC3 (R/O)
OVF_PC2 (R/O)
OVF_PC1 (R/O)
OVF_PC0 (R/O)
Reserved
RESET Value — 0x00000000_00000000
Figure 30-19. Layout of MSR_UNCORE_PERF_GLOBAL_STATUS MSR
Figure 30-20 shows the layout of MSR_UNCORE_PERF_GLOBAL_OVF_CTRL.
63 62 61 60
32 31
8 7 6 5 43 2 1 0
CLR_CHG (WO1)
CLR_OVF_PMI (WO1)
CLR_OVF_FC0 (WO1)
CLR_OVF_PC7 (WO1)
CLR_OVF_PC6 (WO1)
CLR_OVF_PC5 (WO1)
CLR_OVF_PC4 (WO1)
CLR_OVF_PC3 (WO1)
CLR_OVF_PC2 (WO1)
CLR_OVF_PC1 (WO1)
CLR_OVF_PC0 (WO1)
Reserved
RESET Value — 0x00000000_00000000
Figure 30-20. Layout of MSR_UNCORE_PERF_GLOBAL_OVF_CTRL MSR
Vol. 3B 30-41
PERFORMANCE MONITORING
•
CLR_OVF_PCn (bit n, n = 0, 7): Set this bit to clear the overflow status for
general-purpose uncore counter MSR_UNCORE_PerfCntr n. Writing a value other
than 1 is ignored.
•
CLR_OVF_FC0 (bit 32): Set this bit to clear the overflow status for the fixedfunction uncore counter MSR_UNCORE_FixedCntr0. Writing a value other than 1
is ignored.
•
CLR_OVF_PMI (bit 61): Set this bit to clear the OVF_PMI flag in
MSR_UNCORE_PERF_GLOBAL_STATUS. Writing a value other than 1 is ignored.
•
CLR_CHG (bit 63): Set this bit to clear the CHG flag in
MSR_UNCORE_PERF_GLOBAL_STATUS register. Writing a value other than 1 is
ignored.
30.6.2.2
Uncore Performance Event Configuration Facility
MSR_UNCORE_PerfEvtSel0 through MSR_UNCORE_PerfEvtSel7 are used to select
performance event and configure the counting behavior of the respective uncore
performance counter. Each uncore PerfEvtSel MSR is paired with an uncore performance counter. Each uncore counter must be locally configured using the corresponding MSR_UNCORE_PerfEvtSelx and counting must be enabled using the
respective EN_PCx bit in MSR_UNCORE_PERF_GLOBAL_CTRL. Figure 30-21 shows
the layout of MSR_UNCORE_PERFEVTSELx.
63
31
24 23 22 21 20 19 18 17 16 15
Counter Mask
(CMASK)
0
8 7
Unit Mask (UMASK)
Event Select
INV—Invert counter mask
EN—Enable counters
PMI—Enable PMI on overflow
E—Edge detect
OCC_CTR_RST—Rest Queue Occ
Reserved
RESET Value — 0x00000000_00000000
Figure 30-21. Layout of MSR_UNCORE_PERFEVTSELx MSRs
•
•
Event Select (bits 7:0): Selects the event logic unit used to detect uncore events.
•
OCC_CTR_RST (bit17): When set causes the queue occupancy counter
associated with this event to be cleared (zeroed). Writing a zero to this bit will be
ignored. It will always read as a zero.
Unit Mask (bits 15:8) : Condition qualifiers for the event selection logic specified
in the Event Select field.
30-42 Vol. 3B
PERFORMANCE MONITORING
•
Edge Detect (bit 18): When set causes the counter to increment when a
deasserted to asserted transition occurs for the conditions that can be expressed
by any of the fields in this register.
•
PMI (bit 20): When set, the uncore will generate an interrupt request when this
counter overflowed. This request will be routed to the logical processors as
enabled in the PMI enable bits (EN_PMI_COREx) in the register
MSR_UNCORE_PERF_GLOBAL_CTRL.
•
EN (bit 22): When clear, this counter is locally disabled. When set, this counter is
locally enabled and counting starts when the corresponding EN_PCx bit in
MSR_UNCORE_PERF_GLOBAL_CTRL is set.
•
INV (bit 23): When clear, the Counter Mask field is interpreted as greater than or
equal to. When set, the Counter Mask field is interpreted as less than.
•
Counter Mask (bits 31:24): When this field is clear, it has no effect on counting.
When set to a value other than zero, the logical processor compares this field to
the event counts on each core clock cycle. If INV is clear and the event counts are
greater than or equal to this field, the counter is incremented by one. If INV is set
and the event counts are less than this field, the counter is incremented by one.
Otherwise the counter is not incremented.
Figure 30-22 shows the layout of MSR_UNCORE_FIXED_CTR_CTRL.
63
8 7 6 5 43 2 1 0
PMI - Generate PMI on overflow
EN - Enable
Reserved
RESET Value — 0x00000000_00000000
Figure 30-22. Layout of MSR_UNCORE_FIXED_CTR_CTRL MSR
•
EN (bit 0): When clear, the uncore fixed-function counter is locally disabled.
When set, it is locally enabled and counting starts when the EN_FC0 bit in
MSR_UNCORE_PERF_GLOBAL_CTRL is set.
•
PMI (bit 2): When set, the uncore will generate an interrupt request when the
uncore fixed-function counter overflowed. This request will be routed to the
logical processors as enabled in the PMI enable bits (EN_PMI_COREx) in the
register MSR_UNCORE_PERF_GLOBAL_CTRL.
Both the general-purpose counters (MSR_UNCORE_PerfCntr) and the fixed-function
counter (MSR_UNCORE_FixedCntr0) are 48 bits wide. They support both counting
Vol. 3B 30-43
PERFORMANCE MONITORING
and sampling usages. The event logic unit can filter event counts to specific regions
of code or transaction types incoming to the home node logic.
30.6.2.3
Uncore Address/Opcode Match MSR
The Event Select field [7:0] of MSR_UNCORE_PERFEVTSELx is used to select
different uncore event logic unit. When the event “ADDR_OPCODE_MATCH“ is
selected in the Event Select field, software can filter uncore performance events
according to transaction address and certain transaction responses. The address
filter and transaction response filtering requires the use of
MSR_UNCORE_ADDR_OPCODE_MATCH register. The layout is shown in
Figure 30-23.
63
60
48 47
40 39
3 2 0
ADDR
Opcode
MatchSel—Select addr/Opcode
Opcode—Opcode and Message
ADDR—Bits 39:4 of physical address
Reserved
RESET Value — 0x00000000_00000000
Figure 30-23. Layout of MSR_UNCORE_ADDR_OPCODE_MATCH MSR
•
Addr (bits 39:3): The physical address to match if “MatchSel“ field is set to select
address match. The uncore performance counter will increment if the lowest 40bit incoming physical address (excluding bits 2:0) for a transaction request
matches bits 39:3.
•
Opcode (bits 47:40) : Bits 47:40 allow software to filter uncore transactions
based on QPI link message class/packed header opcode. These bits are consists
two sub-fields:
— Bits 43:40 specify the QPI packet header opcode,
— Bits 47:44 specify the QPI message classes.
Table 30-16 lists the encodings supported in the opcode field.
Table 30-16. Opcode Field Encoding for MSR_UNCORE_ADDR_OPCODE_MATCH
Opcode [43:40]
30-44 Vol. 3B
QPI Message Class
Home Request
Snoop Response
Data Response
[47:44] = 0000B
[47:44] = 0001B
[47:44] = 1110B
PERFORMANCE MONITORING
Table 30-16. Opcode Field Encoding for MSR_UNCORE_ADDR_OPCODE_MATCH
Opcode [43:40]
QPI Message Class
1
DMND_IFETCH
2
2
WB
3
3
PF_DATA_RD
4
4
PF_RFO
5
5
PF_IFETCH
6
6
OTHER
7
7
NON_DRAM
15
15
•
MatchSel (bits 63:61): Software specifies the match criteria according to the
following encoding:
— 000B: Disable addr_opcode match hardware
— 100B: Count if only the address field matches,
— 010B: Count if only the opcode field matches
— 110B: Count if either opcode field matches or the address field matches
— 001B: Count only if both opcode and address field match
— Other encoding are reserved
30.6.3
Intel Xeon Processor 7500 Series Performance Monitoring
Facility
The performance monitoring facility in the processor core of Intel Xeon processor
7500 series are the same as those supported in Intel Xeon processor 5500 series.
The uncore subsystem in Intel Xeon processor 7500 series are significantly different
The uncore performance monitoring facility consist of many distributed units associated with individual logic control units (referred to as boxes) within the uncore
subsystem. A high level block diagram of the various box units of the uncore is shown
in Figure 30-24.
Uncore PMUs are programmed via MSR interfaces. Each of the distributed uncore
PMU units have several general-purpose counters. Each counter requires an associated event select MSR, and may require additional MSRs to configure sub-event
conditions. The uncore PMU MSRs associated with each box can be categorized based
on its functional scope: per-counter, per-box, or global across the uncore. The
number counters available in each box type are different. Each box generally
provides a set of MSRs to enable/disable, check status/overflow of multiple counters
within each box.
Vol. 3B 30-45
PERFORMANCE MONITORING
L3 Cache
CBox
CBox
CBox
CBox
CBox
SBox
CBox
CBox
CBox
SBox
SMI Channels
PBox
MBox
BBox
RBox
BBox
MBox
PBox
SMI Channels
WBox
PBox
PBox
PBox
PBox
UBox
4 Intel QPI Links
Figure 30-24. Distributed Units of the Uncore of Intel Xeon Processor 7500 Series
Table 30-17 summarizes the number MSRs for uncore PMU for each box.
Table 30-17. Uncore PMU MSR Summary
Box
# of
Boxes
Counters per Box
Counter
Width
General
Purpose
Global
Enable
Sub-control MSRs
C-Box
8
6
48
Yes
per-box
None
S-Box
2
4
48
Yes
per-box
Match/Mask
B-Box
2
4
48
Yes
per-box
Match/Mask
M-Box
2
6
48
Yes
per-box
Yes
R-Box
1
16 ( 2 port, 8 per
port)
48
Yes
per-box
Yes
W-Box
1
4
48
Yes
per-box
None
1
48
No
per-box
None
1
48
Yes
uncore
None
U-Box
1
30-46 Vol. 3B
PERFORMANCE MONITORING
The W-Box provides 4 general-purpose counters, each requiring an event select
configuration MSR, similar to the general-purpose counters in other boxes. There is
also a fixed-function counter that increments clockticks in the uncore clock domain.
For C,S,B,M,R, and W boxes, each box provides an MSR to enable/disable counting,
configuring PMI of multiple counters within the same box, this is somewhat similar
the “global control“ programming interface, IA32_PERF_GLOBAL_CTRL, offered in
the core PMU. Similarly status information and counter overflow control for multiple
counters within the same box are also provided in C,S,B,M,R, and W boxes.
In the U-Box, MSR_U_PMON_GLOBAL_CTL provides overall uncore PMU
enable/disable and PMI configuration control. The scope of status information in the
U-box is at per-box granularity, in contrast to the per-box status information MSR (in
the C,S,B,M,R, and W boxes) providing status information of individual counter overflow. The difference in scope also apply to the overflow control MSR in the U-Box
versus those in the other Boxes.
The individual MSRs that provide uncore PMU interfaces are listed in Appendix B.
Table B-7 under the general naming style of
MSR_%box#%_PMON_%scope_function%, where %box#% designates the type of
box and zero-based index if there are more the one box of the same type,
%scope_function% follows the examples below:
•
Multi-counter enabling MSRs: MSR_U_PMON_GLOBAL_CTL,
MSR_S0_PMON_BOX_CTL, MSR_C7_PMON_BOX_CTL, etc.
•
Multi-counter status MSRs: MSR_U_PMON_GLOBAL_STATUS,
MSR_S0_PMON_BOX_STATUS, MSR_C7_PMON_BOX_STATUS, etc.
•
Multi-counter overflow control MSRs: MSR_U_PMON_GLOBAL_OVF_CTL,
MSR_S0_PMON_BOX_OVF_CTL, MSR_C7_PMON_BOX_OVF_CTL, etc.
•
Performance counters MSRs: the scope is implicitly per counter, e.g.
MSR_U_PMON_CTR, MSR_S0_PMON_CTR0, MSR_C7_PMON_CTR5, etc
•
Event select MSRs: the scope is implicitly per counter, e.g.
MSR_U_PMON_EVNT_SEL, MSR_S0_PMON_EVNT_SEL0,
MSR_C7_PMON_EVNT_SEL5, etc
•
Sub-control MSRs: the scope is implicitly per-box granularity, e.g.
MSR_M0_PMON_TIMESTAMP, MSR_R0_PMON_IPERF0_P1, MSR_S1_PMON_MATCH.
Details of uncore PMU MSR bit field definitions can be found in a separate document
“Intel Xeon Processor 7500 Series Uncore Performance Monitoring Guide“.
Vol. 3B 30-47
PERFORMANCE MONITORING
30.7
PERFORMANCE MONITORING FOR PROCESSORS
BASED ON INTEL® MICROARCHITECTURE CODE
NAME WESTMERE
All of the performance monitoring programming interfaces (architectural and nonarchitectural core PMU facilities, and uncore PMU) described in Section 30.6 also
apply to processors based on Intel® microarchitecture code name Westmere.
Table 30-14 describes a non-architectural performance monitoring event (event code
0B7H) and associated MSR_OFFCORE_RSP_0 (address 1A6H) in the core PMU. This
event and a second functionally equivalent offcore response event using event code
0BBH and MSR_OFFCORE_RSP_1 (address 1A7H) are supported in processors based
on Intel microarchitecture code name Westmere. The event code and event mask
definitions of Non-architectural performance monitoring events are listed in Table
A-11.
The load latency facility is the same as described in Section 30.6.1.2, but added
enhancement to provide more information in the data source encoding field of each
load latency record. The additional information relates to STLB_MISS and LOCK, see
Table 30-22.
30.7.1
Intel Xeon Processor E7 Family Performance Monitoring
Facility
The performance monitoring facility in the processor core of the Intel Xeon processor
E7 family is the same as those supported in the Intel Xeon processor 5600 series2.
The uncore subsystem in the Intel Xeon processor E7 family is similar to those of the
Intel Xeon processor 7500 series. The high level construction of the uncore subsystem is similar to that shown in Figure 30-24, with the additional capability that up
to 10 C-Box units are supported.
Table 30-18 summarizes the number MSRs for uncore PMU for each box.
Table 30-18. Uncore PMU MSR Summary for Intel Xeon Processor E7 Family
Box
# of
Boxes
Counters per Box
Counter
Width
General
Purpose
Global
Enable
Sub-control MSRs
C-Box
10
6
48
Yes
per-box
None
S-Box
2
4
48
Yes
per-box
Match/Mask
B-Box
2
4
48
Yes
per-box
Match/Mask
M-Box
2
6
48
Yes
per-box
Yes
2. Exceptions are indicated for event code 0FH in .Table A-6; and valid bits of data source
encoding field of each load latency record is limited to bits 5:4 of Table 30-22.
30-48 Vol. 3B
PERFORMANCE MONITORING
Table 30-18. Uncore PMU MSR Summary for Intel Xeon Processor E7 Family
Box
# of
Boxes
R-Box
1
W-Box
1
U-Box
1
30.8
Counter
Width
General
Purpose
Global
Enable
Sub-control MSRs
16 ( 2 port, 8 per
port)
48
Yes
per-box
Yes
4
48
Yes
per-box
None
1
48
No
per-box
None
1
48
Yes
uncore
None
Counters per Box
PERFORMANCE MONITORING FOR PROCESSORS
BASED ON INTEL® MICROARCHITECTURE CODE
NAME SANDY BRIDGE
Intel Core i7, i5, i3 processors 2xxx series are based on Intel microarchitecture code
name Sandy Bridge, this section describes the performance monitoring facilities
provided in the processor core. The core PMU supports architectural performance
monitoring capability with version ID 3 (see Section 30.2.2.2) and a host of nonarchitectural monitoring capabilities.
Architectural performance monitoring events and non-architectural monitoring
events are programmed using fixed counters and programmable counters/event
select MSRS described in Section 30.2.2.2.
The core PMU’s capability is similar to those described in Section 30.6.1 and Section
30.7, with some differences and enhancements relative to Intel microarchitecture
code name Westmere summarized in Table 30-19.
Table 30-19. Core PMU Comparison
Box
Sandy Bridge
Westmere
Comment
# of Fixed counters
per thread
3
3
Use CPUID to enumerate
# of counters
# of general-purpose
counters per core
8
8
Counter width (R,W)
R:48 , W: 32/48
R:48, W:32
see Section 30.2.2.3
# of programmable
counters per thread
4 or (8 if a core not shared
by two threads)
4
Use CPUID to enumerate
# of counters
PEBS Events
See Table 30-21
See Table 30-10
IA32_PMC4-IA32_PMC7
do not support PEBS.
Vol. 3B 30-49
PERFORMANCE MONITORING
Table 30-19. Core PMU Comparison
Box
Sandy Bridge
Westmere
PEBS-Load Latency
Data source/ STLB/Lock
encoding; See Section
30.8.4.2
Data source
encoding
PEBS-Precise Store
Section 30.8.4.3
No
PEBS-PDIR
yes (using precise
INST_RETIRED.ALL)
No PDIR, no
INST_RETIRED.ALL
Off-core Response
Event
MSR 1A6H and 1A7H;
Extended request and
response types
MSR 1A6H and
1A7H, limited
types
30.8.1
Comment
Nehalem supports 1A6H
only.
Global Counter Control Facilities In Intel® microarchitecture
code name Sandy Bridge
The number of general-purpose performance counters visible to a logical processor
can vary across Processors based on Intel microarchitecture code name Sandy
Bridge. Software must use CPUID to determine the number performance
counters/event select registers (See Section 30.2.1.1).
63
35 34 33 32 31
8 7 6 5 4 3 2 1 0
FIXED_CTR2 enable
FIXED_CTR1 enable
FIXED_CTR0 enable
PMC7_EN (if PMC7 present)
PMC6_EN (if PMC6 present)
PMC5_EN (if PMC5 present)
PMC4_EN (if PMC4 present)
PMC3_EN
PMC2_EN
PMC1_EN
PMC0_EN
Reserved
Valid if CPUID.0AH:EAX[15:8] = 8, else reserved.
Figure 30-25. IA32_PERF_GLOBAL_CTRL MSR in Intel microarchitecture code name
Sandy Bridge
Figure 30-10 depicts the layout of IA32_PERF_GLOBAL_CTRL MSR. The enable bits
(PMC4_EN, PMC5_EN, PMC6_EN, PMC7_EN) corresponding to IA32_PMC4-
30-50 Vol. 3B
PERFORMANCE MONITORING
IA32_PMC7 are valid only if CPUID.0AH:EAX[15:8] reports a value of ‘8’. If
CPUID.0AH:EAX[15:8] = 4, attempts to set the invalid bits will cause #GP.
Each enable bit in IA32_PERF_GLOBAL_CTRL is AND’ed with the enable bits for all
privilege levels in the respective IA32_PERFEVTSELx or
IA32_PERF_FIXED_CTR_CTRL MSRs to start/stop the counting of respective
counters. Counting is enabled if the AND’ed results is true; counting is disabled when
the result is false.
IA32_PERF_GLOBAL_STATUS MSR provides single-bit status used by software to
query the overflow condition of each performance counter. The MSR also provides
additional status bit to indicate overflow conditions when counters are programmed
for precise-event-based sampling (PEBS). The IA32_PERF_GLOBAL_STATUS MSR
also provides a ‘sticky bit’ to indicate changes to the state of performance monitoring
hardware (see Figure 30-26). A value of 1 in each bit of the PMCx_OVF field indicates
an overflow condition has occurred in the associated counter.
63 62
35 34 33 32 31
8 7 6 5 4 3 2 1 0
CondChgd
OvfBuffer
FIXED_CTR2 Overflow
FIXED_CTR1 Overflow
FIXED_CTR0 Overflow
PMC7_OVF (If PMC7 present)
PMC6_OVF (If PMC6 present)
PMC5_OVF (If PMC5 present)
PMC4_OVF (If PMC4 present)
PMC3_OVF
PMC2_OVF
PMC1_OVF
PMC0_OVF
Reserved
Valid if CPUID.0AH:EAX[15:8] = 8; else reserved
Figure 30-26. IA32_PERF_GLOBAL_STATUS MSR in Intel microarchitecture code
name Sandy Bridge
When a performance counter is configured for PEBS, an overflow condition in the
counter generates a performance-monitoring interrupt this signals a PEBS event. On
a PEBS event, the processor stores data records in the buffer area (see Section
16.4.9), clears the counter overflow status, and sets the OvfBuffer bit in
IA32_PERF_GLOBAL_STATUS.
IA32_PERF_GLOBAL_OVF_CTL MSR allows software to clear overflow the indicators
for general-purpose or fixed-function counters via a single WRMSR (see
Figure 30-27). Clear overflow indications when:
Vol. 3B 30-51
PERFORMANCE MONITORING
•
Setting up new values in the event select and/or UMASK field for counting or
sampling
•
•
Reloading counter values to continue sampling
Disabling event counting or sampling
63 62
35 34 33 32 31
8 7 6 5 4 3
2 1 0
ClrCondChgd
ClrOvfBuffer
FIXED_CTR2 ClrOverflow
FIXED_CTR1 ClrOverflow
FIXED_CTR0 ClrOverflow
PMC7_ClrOvf (if PMC7 present)
PMC6_ClrOvf (if PMC6 present)
PMC5_ClrOvf (if PMC5 present)
PMC4_ClrOvf (if PMC4 present)
PMC3_ClrOvf
PMC2_ClrOvf
PMC1_ClrOvf
PMC0_ClrOvf
Reserved
Valid if CPUID.0AH:EAX[15:8] = 8; else reserved
Figure 30-27. IA32_PERF_GLOBAL_OVF_CTRL MSR in Intel microarchitecture code
name Sandy Bridge
30.8.2
Counter Coalescence
In processors based on Intel microarchitecture code name Sandy Bridge, each
processor core implements eight general-purpose counters. CPUID.0AH:EAX[15:8]
will report either 4 or 8 depending specific processor’s product features.
If a processor core is shared by two logical processors, each logical processors can
access 4 counters (IA32_PMC0-IA32_PMC3). This is the same as in the prior generation for processors based on Intel microarchitecture code name Nehalem.
If a processor core is not shared by two logical processors, all eight general-purpose
counters are visible, and CPUID.0AH:EAX[15:8] reports 8. IA32_PMC4-IA32_PMC7
occupy MSR addresses 0C5H through 0C8H. Each counter is accompanied by an
event select MSR (IA32_PERFEVTSEL4-IA32_PERFEVTSEL7).
If CPUID.0AH:EAX[15:8] report 4, access to IA32_PMC4-IA32_PMC7, IA32_PMC4IA32_PMC7 will cause #GP. Writing 1’s to bit position 7:4 of
IA32_PERF_GLOBAL_CTRL, IA32_PERF_GLOBAL_STATUS, or
IA32_PERF_GLOBAL_OVF_CTL will also cause #GP.
30-52 Vol. 3B
PERFORMANCE MONITORING
30.8.3
Full Width Writes to Performance Counters
Processors based on Intel microarchitecture code name Sandy Bridge support fullwidth writes to the general-purpose counters, IA32_PMCx. Support of full-width
writes are enumerated by IA32_PERF_CAPABILITIES.FW_WRITES[13] (see Section
30.2.2.3).
The default behavior of IA32_PMCx is unchanged, i.e. WRMSR to IA32_PMCx results
in a sign-extended 32-bit value of the input EAX written into IA32_PMCx. Full-width
writes must issue WRMSR to a dedicated alias MSR address for each IA32_PMCx.
Software must check the presence of full-width write capability and the presence of
the alias address IA32_A_PMCx by testing IA32_PERF_CAPABILITIES[13].
30.8.4
PEBS Support in Intel® microarchitecture code name Sandy
Bridge
Processors based on Intel microarchitecture code name Sandy Bridge support PEBS,
similar to those offered in prior generation, with several enhanced features. The key
components and differences of PEBS facility relative to Intel microarchitecture code
name Westmere is summarized in Table 30-20.
Table 30-20. PEBS Facility Comparison
Box
Sandy Bridge
Westmere
Comment
Valid IA32_PMCx
PMC0-PMC3
PMC0-PMC3
No PEBS on PMC4-PMC7
PEBS Buffer
Programming
Section 30.6.1.1
Section 30.6.1.1
Unchanged
IA32_PEBS_ENABLE
Layout
Figure 30-28
Figure 30-14
PEBS record layout
Physical Layout same
as Table 30-12
Table 30-12
Enhanced fields at
offsets 98H, A0H, A8H
PEBS Events
See Table 30-21
See Table 30-10
IA32_PMC4-IA32_PMC7
do not support PEBS.
PEBS-Load Latency
See Table 30-22
Table 30-13
PEBS-Precise Store
yes; see Section
30.8.4.3
No
IA32_PMC3 only
PEBS-PDIR
yes
No
IA32_PMC1 only
SAMPLING
Restriction
Small SAV(CountDown) value incur higher
overhead than prior generation.
Only IA32_PMC0 through IA32_PMC3 support PEBS.
Vol. 3B 30-53
PERFORMANCE MONITORING
NOTE
PEBS events are only valid when the following fields of
IA32_PERFEVTSELx are all zero: AnyThread, Edge, Invert, CMask.
In IA32_PEBS_ENABLE MSR, bit 63 is defined as PS_ENABLE: When set, this enables
IA32_PMC3 to capture precise store information. Only IA32_PMC3 supports the
precise store facility.
36 3534 33 32 31
63 62
8 7 6 5 43 2 1 0
PS_EN (R/W)
LL_EN_PMC3 (R/W)
LL_EN_PMC2 (R/W)
LL_EN_PMC1 (R/W)
LL_EN_PMC0 (R/W)
PEBS_EN_PMC3 (R/W)
PEBS_EN_PMC2 (R/W)
PEBS_EN_PMC1 (R/W)
PEBS_EN_PMC0 (R/W)
Reserved
RESET Value — 0x00000000_00000000
Figure 30-28. Layout of IA32_PEBS_ENABLE MSR
30.8.4.1
PEBS Record Format
The layout of PEBS records physically identical to those shown in Table 30-12, but the
fields at offset 98H, A0H and A8H have been enhanced to support additional PEBS
capabilities.
•
Load/Store Data Linear Address (Offset 98H): This field will contain the linear
address of the source of the load, or linear address of the destination of the store.
•
Data Source /Store Status (Offset A0H):When load latency is enabled, this field
will contain three piece of information (including an encoded value indicating the
source which satisfied the load operation). The source field encodings are
detailed in Table 30-13. When precise store is enabled, this field will contain
information indicating the status of the store, as detailed in Table 19.
•
Latency Value/0 (Offset A8H): When load latency is enabled, this field contains
the latency in cycles to service the load. This field is not meaningful when precise
store is enabled and will be written to zero in that case. Upon writing the PEBS
record, microcode clears the overflow status bits in the
IA32_PERF_GLOBAL_STATUS corresponding to those counters that both
30-54 Vol. 3B
PERFORMANCE MONITORING
overflowed and were enabled in the IA32_PEBS_ENABLE register. The status bits
of other counters remain unaffected.
The number PEBS events has expanded. The list of PEBS events supported in Intel
microarchitecture code name Sandy Bridge is shown in Table 30-21.
Table 30-21. PEBS Performance Events for Intel microarchitecture code name Sandy
Bridge
Event Name
Event Select Sub-event
UMask
INST_RETIRED
C0H
PREC_DIST
01H1
UOPS_RETIRED
C2H
All
01H
Retire_Slots
02H
BR_INST_RETIRED
BR_MISP_RETIRED
MEM_TRANS_RETIRED
MEM_UOP_RETIRED
MEM_LOAD_UOPS_RETIRED
C4H
C5H
CDH
D0H
D1H
Conditional
01H
Near_Call
02H
All_branches
04H
Near_Return
08H
Not_Taken
10H
Near_Taken
20H
Far_Branches
40H
Conditional
01H
Near_Call
02H
All_branches
04H
Not_Taken
10H
Taken
20H
Load_Latency
01H
Precise_Store
02H
Load
01H
Store
02H
STLB_Miss
10H
Lock
20H
SPLIT
40H
ALL
80H
L1_Hit
01H
L2_Hit
02H
L3_Hit
04H
Hit_LFB
40H
Vol. 3B 30-55
PERFORMANCE MONITORING
Table 30-21. PEBS Performance Events for Intel microarchitecture (Contd.)code name
Sandy Bridge
Event Name
Event Select Sub-event
UMask
MEM_LOAD_UOPS_LLC_HIT_RETIRED
D2H
01H
MEM_LOAD_UOPS_MISC_RETIRED
D4H
XSNP_Miss
XSNP_Hit
02H
XSNP_Hitm
04H
XSNP_None
08H
LLC_Miss
02H
NOTES:
1. Only available on IA32_PMC1.
30.8.4.2
Load Latency Performance Monitoring Facility
The load latency facility in Intel microarchitecture code name Sandy Bridge is similar
to that in prior microarchitecture. It provides software a means to characterize the
average load latency to different levels of cache/memory hierarchy. This facility
requires processor supporting enhanced PEBS record format in the PEBS buffer, see
Table 30-12 and Section 30.8.4.1. The facility measures latency from micro-operation (uop) dispatch to when data is globally observable (GO).
To use this feature software must assure:
•
One of the IA32_PERFEVTSELx MSR is programmed to specify the event unit
MEM_TRANS_RETIRED, and the LATENCY_ABOVE_THRESHOLD event mask must be
specified (IA32_PerfEvtSelX[15:0] = 0x1CDH). The corresponding counter
IA32_PMCx will accumulate event counts for architecturally visible loads which
exceed the programmed latency threshold specified separately in a MSR. Stores
are ignored when this event is programmed. The CMASK or INV fields of the
IA32_PerfEvtSelX register used for counting load latency must be 0. Writing
other values will result in undefined behavior.
•
The MSR_PEBS_LD_LAT_THRESHOLD MSR is programmed with the desired
latency threshold in core clock cycles. Loads with latencies greater than this
value are eligible for counting and latency data reporting. The minimum value
that may be programmed in this register is 3 (the minimum detectable load
latency is 4 core clock cycles).
•
The PEBS enable bit in the IA32_PEBS_ENABLE register is set for the corresponding IA32_PMCx counter register. This means that both the PEBS_EN_CTRX
and LL_EN_CTRX bits must be set for the counter(s) of interest. For example, to
enable load latency on counter IA32_PMC0, the IA32_PEBS_ENABLE register
must be programmed with the 64-bit value 0x00000001.00000001.
•
When Load latency event is enabled, no other PEBS event can be configured with
other counters.
30-56 Vol. 3B
PERFORMANCE MONITORING
When the load-latency facility is enabled, load operations are randomly selected by
hardware and tagged to carry information related to data source locality and latency.
Latency and data source information of tagged loads are updated internally. The
MEM_TRANS_RETIRED event for load latency counts only tagged retired loads. If a
load is cancelled it will not be counted and the internal state of the load latency
facility will not be updated. In this case the hardware will tag the next available load.
When a PEBS assist occurs, the last update of latency and data source information
are captured by the assist and written as part of the PEBS record. The PEBS sample
after value (SAV), specified in PEBS CounterX Reset, operates orthogonally to the
tagging mechanism. Loads are randomly tagged to collect latency data. The SAV
controls the number of tagged loads with latency information that will be written into
the PEBS record field by the PEBS assists. The load latency data written to the PEBS
record will be for the last tagged load operation which retired just before the PEBS
assist was invoked.
The physical layout of the PEBS records is the same as shown in Table 30-12. The
specificity of Data Source entry at offset A0H has been enhanced to report three
piece of information.
Table 30-22. Layout of Data Source Field of Load Latency Record
Field
Position
Description
Source
3:0
See Table 30-13
STLB_MISS 4
0: The load did not miss the STLB (hit the DTLB or STLB).
1: The load missed the STLB.
Lock
5
0: The load was not part of a locked transaction.
1: The load was part of a locked transaction.
Reserved
63:6
The layout of MSR_PEBS_LD_LAT_THRESHOLD is the same as shown in
Figure 30-16.
30.8.4.3
Precise Store Facility
Processors based on Intel microarchitecture code name Sandy Bridge offer a precise
store capability that complements the load latency facility. It provides a means to
profile store memory references in the system.
Precise stores leverage the PEBS facility and provide additional information about
sampled stores. Having precise memory reference events with linear address information for both loads and stores can help programmers improve data structure
layout, eliminate remote node references, and identify cache-line conflicts in NUMA
systems.
Vol. 3B 30-57
PERFORMANCE MONITORING
Only IA32_PMC3 can be used to capture precise store information. After enabling this
facility, counter overflows will initiate the generation of PEBS records as previously
described in PEBS. Upon counter overflow hardware captures the linear address and
other status information of the next store that retires. This information is then
written to the PEBS record.
To enable the precise store facility, software must complete the following steps.
Please note that the precise store facility relies on the PEBS facility, so the PEBS
configuration requirements must be completed before attempting to capture precise
store information.
•
•
Complete the PEBS configuration steps.
•
Set IA32_PEBS_ENABLE[3] and IA32_PEBS_ENABLE[63]. This enables
IA32_PMC3 as a PEBS counter and enables the precise store facility, respectively.
Program the MEM_TRANS_RETIRED.PRECISE_STORE event in
IA32_PERFEVTSEL3. Only counter 3 (IA32_PMC3) supports collection of precise
store information.
The precise store information written into a PEBS record affects entries at offset 98H,
A0H and A8H of Table 30-12. The specificity of Data Source entry at offset A0H has
been enhanced to report three piece of information.
Table 30-23. Layout of Precise Store Information In PEBS Record
Field
Offset
Description
Store Data
Linear Address
98H
The linear address of the destination of the store.
Store Status
A0H
DCU Hit (Bit 0): The store hit the data cache closest to the core (lowest
latency cache) if this bit is set, otherwise the store missed the data
cache.
STLB Miss (bit 4): The store missed the STLB if set, otherwise the store
hit the STLB
Locked Access (bit 5): The store was part of a locked access if set,
otherwise the store was not part of a locked access.
Reserved
30.8.4.4
A8H
Reserved
Precise Distribution of Instructions Retired (PDIR)
Upon triggering a PEBS assist, there will be a finite delay between the time the
counter overflows and when the microcode starts to carry out its data collection obligations. INST_RETIRED is a very common event that is used to sample where performance bottleneck happened and to help identify its location in instruction address
space. Even if the delay is constant in core clock space, it invariably manifest as variable “skids” in instruction address space. This creates a challenge for programmers
to profile a workload and pinpoint the location of bottlenecks.
30-58 Vol. 3B
PERFORMANCE MONITORING
The core PMU in processors based on Intel microarchitecture code name Sandy
Bridge include a facility referred to as precise distribution of Instruction Retired
(PDIR).
The PDIR facility mitigates the “skid“ problem by providing an early indication of
when the INST_RETIRED counter is about to overflow, allowing the machine to more
precisely trap on the instruction that actually caused the counter overflow thus eliminating skid.
PDIR applies only to the INST_RETIRED.PREC_DIST precise event, and must use
IA32_PMC1 with PerfEvtSel1 property configured and bit 1 in the
IA32_PEBS_ENABLE set to 1. INST_RETIRED.PREC_DIST is a non-architectural
performance event, it is not supported in prior generation microarchitectures. Additionally, current implementation of PDIR limits tool to quiesce the rest of the
programmable counters in the core when PDIR is active.
30.8.5
Off-core Response Performance Monitoring
The core PMU in processors based on Intel microarchitecture code name Sandy
Bridge provides off-core response facility similar to prior generation. Off-core
response can be programed only with a specific pair of event select and counter MSR,
and with specific event codes and predefine mask bit value in a dedicated MSR to
specify attributes of the off-core transaction. Two event codes are dedicated for offcore response event programming. Each event code for off-core response monitoring
requires programming an associated configuration MSR, MSR_OFFCORE_RSP_x.
Table 30-24 lists the event code, mask value and additional off-core configuration
MSR that must be programmed to count off-core response events using IA32_PMCx.
Table 30-24. Off-Core Response Event Encoding
Counter
Event code
UMask
Required Off-core Response MSR
PMC0
0xB7
0x01
MSR_OFFCORE_RSP_0 (address 0x1A6)
PMC3
0xBB
0x01
MSR_OFFCORE_RSP_1 (address 0x1A7)
The layout of MSR_OFFCORE_RSP_0 and MSR_OFFCORE_RSP_1 are shown in
Figure 30-29 and Figure 30-30. Bits 15:0 specifies the request type of a transaction
request to the uncore. Bits 30:16 specifies supplier information, bits 37:31 specifies
snoop response information.
Vol. 3B 30-59
PERFORMANCE MONITORING
63
37
15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
See Figure 3-30
RESPONSE TYPE — Other (R/W)
RESERVED
REQUEST TYPE — STRM_ST (R/W)
REQUEST TYPE — BUS_LOCKS (R/W)
REQUEST TYPE — PF_LLC_IFETCH (R/W)
REQUEST TYPE — PF_LLC_RFO (R/W)
REQUEST TYPE — PF_LLC_DATA_RD (R/W)
REQUEST TYPE — PF_IFETCH (R/W)
REQUEST TYPE — PF_RFO (R/W)
REQUEST TYPE — PF_DATA_RD (R/W)
REQUEST TYPE — WB (R/W)
REQUEST TYPE — DMND_IFETCH (R/W)
REQUEST TYPE — DMND_RFO (R/W)
REQUEST TYPE — DMND_DATA_RD (R/W)
Reserved
RESET Value — 0x00000000_00000000
Figure 30-29. Request_Type Fields for MSR_OFFCORE_RSP_x
Table 30-25. MSR_OFFCORE_RSP_x Request_Type Field Definition
Bit Name
Offset Description
DMND_DATA_RD
0
(R/W). Counts the number of demand and DCU prefetch data reads of
full and partial cachelines as well as demand data page table entry
cacheline reads. Does not count L2 data read prefetches or
instruction fetches.
DMND_RFO
1
(R/W). Counts the number of demand and DCU prefetch reads for
ownership (RFO) requests generated by a write to data cacheline.
Does not count L2 RFO prefetches.
DMND_IFETCH
2
(R/W). Counts the number of demand and DCU prefetch instruction
cacheline reads. Does not count L2 code read prefetches.
WB
3
(R/W). Counts the number of writeback (modified to exclusive)
transactions.
PF_DATA_RD
4
(R/W). Counts the number of data cacheline reads generated by L2
prefetchers.
PF_RFO
5
(R/W). Counts the number of RFO requests generated by L2
prefetchers.
PF_IFETCH
6
(R/W). Counts the number of code reads generated by L2 prefetchers.
30-60 Vol. 3B
PERFORMANCE MONITORING
Table 30-25. MSR_OFFCORE_RSP_x Request_Type Field Definition (Contd.)
Bit Name
Offset Description
PF_LLC_DATA_RD
7
(R/W). L2 prefetcher to L3 for loads.
PF_LLC_RFO
8
(R/W). RFO requests generated by L2 prefetcher
PF_LLC_IFETCH
9
(R/W). L2 prefetcher to L3 for instruction fetches.
BUS_LOCKS
10
(R/W). Bus lock and split lock requests
STRM_ST
11
(R/W). Streaming store requests
OTHER
15
(R/W). Any other request that crosses IDI, including I/O.
63
15 14 13 12 11 10 9
23 22 212019 18 17 16
RESPONSE TYPE — NON_DRAM (R/W)
RSPNS_SNOOP — HITM (R/W)
RSPNS_SNOOP — HIT_FWD
RSPNS_SNOOP — HIT_NO_FWD (R/W)
RSPNS_SNOOP — SNP_MISS (R/W)
RSPNS_SNOOP — SNP_NOT_NEEDED (R/W)
RSPNS_SNOOP — SNPl_NONE (R/W)
RSPNS_SUPPLIER — RESERVED
RSPNS_SUPPLIER — LLC_HITF (R/W)
RSPNS_SUPPLIER — LLC_HITS (R/W)
RSPNS_SUPPLIER — LLC_HITE (R/W)
RSPNS_SUPPLIER — LLC_HITM (R/W)
RSPNS_SUPPLIER — No_SUPP (R/W)
RSPNS_SUPPLIER — ANY (R/W)
Reserved
RESET Value — 0x00000000_00000000
Figure 30-30. Response_Type Fields for MSR_OFFCORE_RSP_x
To properly program this extra register, software must set at least one request type
bit and a valid response type pattern. Otherwise, the event count reported will be
zero. It is permissible and useful to set multiple request and response type bits in
order to obtain various classes of off-core response events.
Table 30-26. MSR_OFFCORE_RSP_x Response Type Field Definition
Subtype
Bit Name
Offset
Description
Common
Any
16
(R/W). Catch all value for any response types.
Vol. 3B 30-61
PERFORMANCE MONITORING
Table 30-26. MSR_OFFCORE_RSP_x Response Type Field Definition (Contd.)
Subtype
Bit Name
Offset
Description
Supplier
Info
NO_SUPP
17
(R/W). No Supplier Information available
LLC_HITM
18
(R/W). M-state initial lookup stat in L3.
LLC_HITE
19
(R/W). E-state
LLC_HITS
20
(R/W). S-state
LLC_HITF
21
(R/W). F-state
Reserved
30:22
Reserved
SNP_NONE
31
(R/W). No details on snoop-related information
Snoop
Info
SNP_NOT_NEEDED 32
(R/W). No snoop was needed to satisfy the request.
SNP_MISS
(R/W). A snoop was needed and it missed all snooped
caches:
33
-For LLC Hit, ReslHitl was returned by all cores
-For LLC Miss, Rspl was returned by all sockets and data
was returned from DRAM.
SNP_NO_FWD
34
(R/W). A snoop was needed and it hits in at least one
snooped cache. Hit denotes a cache-line was valid before
snoop effect. This includes:
-Snoop Hit w/ Invalidation (LLC Hit, RFO)
-Snoop Hit, Left Shared (LLC Hit/Miss, IFetch/Data_RD)
-Snoop Hit w/ Invalidation and No Forward (LLC Miss, RFO
Hit S)
In the LLC Miss case, data is returned from DRAM.
SNP_FWD
35
(R/W). A snoop was needed and data was forwarded
from a remote socket. This includes:
-Snoop Forward Clean, Left Shared (LLC Hit/Miss,
IFetch/Data_RD/RFT).
HITM
36
(R/W). A snoop was needed and it HitM-ed in local or
remote cache. HitM denotes a cache-line was in modified
state before effect as a results of snoop. This includes:
-Snoop HitM w/ WB (LLC miss, IFetch/Data_RD)
-Snoop Forward Modified w/ Invalidation (LLC Hit/Miss,
RFO)
-Snoop MtoS (LLC Hit, IFetch/Data_RD).
NON_DRAM
30-62 Vol. 3B
37
(R/W). Target was non-DRAM system address. This
includes MMIO transactions.
PERFORMANCE MONITORING
To specify a complete offcore response filter, software must properly program bits in
the request and response type fields. A valid request type must have at least one bit
set in the non-reserved bits of 15:0. A valid response type must be a non-zero value
of the following expression:
ANY | [(‘OR’ of Supplier Info Bits) & (‘OR’ of Snoop Info Bits)]
If “ANY“ bit is set, the supplier and snoop info bits are ignored.
30.8.6
Uncore Performance Monitoring Facilities In Intel® Core i7, i5,
i3 Processors 2xxx Series
The uncore sub-system in Intel Core i7, i5, i3 processors 2xxx Series provides a
unified L3 that can support up to four processor cores. The L3 cache consists multiple
slices, each slice interface with a processor via a coherence engine, referred to as a
C-Box. Each C-Box provides dedicated facility of MSRs to select uncore performance
monitoring events and each C-Box event select MSR is paired with a counter register,
similar in style as those described in Section 30.6.2.2. The layout of the event select
MSRs in the C-Boxes are shown in Figure 30-31.
63
28
24 23 22 21 20 19 18 17 16 15
Counter Mask
(CMASK)
0
8 7
Unit Mask (UMASK)
Event Select
INV—Invert counter mask
EN—Enable counters
PMI—Enable PMI on overflow
E—Edge detect
Reserved
RESET Value — 0x00000000_00000000
Figure 30-31. Layout of MSR_UNC_CBO_N_PERFEVTSELx MSR for C-Box N
At the uncore domain level, there is a master set of control MSRs that centrally
manages all the performance monitoring facility of uncore units. Figure 30-32 shows
the layout of the uncore domain global control
MSR bit 31 of MSR_UNC_PERF_GLOBAL_CTRL provides the capability to freeze all
uncore counters when an overflow condition in a unit counter. When set and upon a
counter overflow, the uncore PMU logic will clear the global enable bit, bit 29.
Vol. 3B 30-63
PERFORMANCE MONITORING
63
32 31 30 29 28
4 3 2 1
0
FREEZE—Freeze counters
PMI—Wake cores on PMI
EN—Enable all uncore counters
Core Select — core 3 select
Core Select — core 2 select
Core Select — core 1select
Core Select — core 0 select
Reserved
RESET Value — 0x00000000_00000000
Figure 30-32. Layout of MSR_UNC_PERF_GLOBAL_CTRL MSR for Uncore
Additionally, there is also a fixed counter, counting uncore clockticks, for the uncore
domain. Table 30-27 summarizes the number MSRs for uncore PMU for each box.
Table 30-27. Uncore PMU MSR Summary
Box
# of
Boxes
Counters per Box
Counter
Width
General
Purpose
Global
Enable
C-Box
Up to 4
2
44
Yes
Per-box
1
48
No
Uncore
NCU
30.8.6.1
Comment
Uncore Performance Monitoring Events
There are certain restrictions on the uncore performance counters in each C-Box.
Specifically,
•
Occupancy events are supported only with counter 0 but not counter 1.
Other uncore C-Box events can be programmed with either counter 0 or 1.
The C-Box uncore performance events described in Table A-3 can collect performance characteristics of transactions initiated by processor core. In that respect,
they are similar to various sub-events in the OFFCORE_RESPONSE family of performance events in the core PMU. Information such as data supplier locality (LLC
HIT/MISS) and snoop responses can be collected via OFFCORE_RESPONSE and qualified on a per-thread basis.
On the other hand, uncore performance event logic can not associate its counts with
the same level of per-thread qualification attributes as the core PMU events can.
Therefore, whenever similar event programming capabilities are available from both
30-64 Vol. 3B
PERFORMANCE MONITORING
core PMU and uncore PMU, the recommendation is that utilizing the core PMU events
may be less affected by artifacts, complex interactions and other factors.
30.9
PERFORMANCE MONITORING (PROCESSORS
BASED ON INTEL NETBURST®
MICROARCHITECTURE)
The performance monitoring mechanism provided in Pentium 4 and Intel Xeon
processors is different from that provided in the P6 family and Pentium processors.
While the general concept of selecting, filtering, counting, and reading performance
events through the WRMSR, RDMSR, and RDPMC instructions is unchanged, the
setup mechanism and MSR layouts are incompatible with the P6 family and Pentium
processor mechanisms. Also, the RDPMC instruction has been enhanced to read the
the additional performance counters provided in the Pentium 4 and Intel Xeon
processors and to allow faster reading of counters.
The event monitoring mechanism provided with the Pentium 4 and Intel Xeon
processors (based on Intel NetBurst microarchitecture) consists of the following facilities:
•
The IA32_MISC_ENABLE MSR, which indicates the availability in an Intel 64 or
IA-32 processor of the performance monitoring and precise event-based
sampling (PEBS) facilities.
•
Event selection control (ESCR) MSRs for selecting events to be monitored with
specific performance counters. The number available differs by family and model
(43 to 45).
•
•
18 performance counter MSRs for counting events.
•
•
•
A debug store (DS) save area in memory for storing PEBS records.
•
The MSR_PEBS_ENABLE MSR, which enables the PEBS facilities and replay
tagging used in at-retirement event counting.
•
A set of predefined events and event metrics that simplify the setting up of the
performance counters to count specific events.
18 counter configuration control (CCCR) MSRs, with one CCCR associated with
each performance counter. CCCRs sets up an associated performance counter for
a specific method of counting.
The IA32_DS_AREA MSR, which establishes the location of the DS save area.
The debug store (DS) feature flag (bit 21) returned by the CPUID instruction,
which indicates the availability of the DS mechanism.
Table 30-28 lists the performance counters and their associated CCCRs, along with
the ESCRs that select events to be counted for each performance counter. Predefined
event metrics and events are listed in Appendix A, “Performance-Monitoring Events.”
Vol. 3B 30-65
PERFORMANCE MONITORING
Table 30-28. Performance Counter MSRs and Associated CCCR and
ESCR MSRs (Pentium 4 and Intel Xeon Processors)
Counter
CCCR
ESCR
Name
No.
Addr
Name
Addr
Name
No. Addr
MSR_BPU_COUNTER0
0
300H
MSR_BPU_CCCR0
360H
MSR_BSU_ESCR0
MSR_FSB_ESCR0
MSR_MOB_ESCR0
MSR_PMH_ESCR0
MSR_BPU_ESCR0
MSR_IS_ESCR0
MSR_ITLB_ESCR0
MSR_IX_ESCR0
7
6
2
4
0
1
3
5
3A0H
3A2H
3AAH
3ACH
3B2H
3B4H
3B6H
3C8H
MSR_BPU_COUNTER1
1
301H
MSR_BPU_CCCR1
361H
MSR_BSU_ESCR0
MSR_FSB_ESCR0
MSR_MOB_ESCR0
MSR_PMH_ESCR0
MSR_BPU_ESCR0
MSR_IS_ESCR0
MSR_ITLB_ESCR0
MSR_IX_ESCR0
7
6
2
4
0
1
3
5
3A0H
3A2H
3AAH
3ACH
3B2H
3B4H
3B6H
3C8H
MSR_BPU_COUNTER2
2
302H
MSR_BPU_CCCR2
362H
MSR_BSU_ESCR1
MSR_FSB_ESCR1
MSR_MOB_ESCR1
MSR_PMH_ESCR1
MSR_BPU_ESCR1
MSR_IS_ESCR1
MSR_ITLB_ESCR1
MSR_IX_ESCR1
7
6
2
4
0
1
3
5
3A1H
3A3H
3ABH
3ADH
3B3H
3B5H
3B7H
3C9H
MSR_BPU_COUNTER3
3
303H
MSR_BPU_CCCR3
363H
MSR_BSU_ESCR1
MSR_FSB_ESCR1
MSR_MOB_ESCR1
MSR_PMH_ESCR1
MSR_BPU_ESCR1
MSR_IS_ESCR1
MSR_ITLB_ESCR1
MSR_IX_ESCR1
7
6
2
4
0
1
3
5
3A1H
3A3H
3ABH
3ADH
3B3H
3B5H
3B7H
3C9H
MSR_MS_COUNTER0
4
304H
MSR_MS_CCCR0
364H
MSR_MS_ESCR0
MSR_TBPU_ESCR0
MSR_TC_ESCR0
0
2
1
3C0H
3C2H
3C4H
MSR_MS_COUNTER1
5
305H
MSR_MS_CCCR1
365H
MSR_MS_ESCR0
MSR_TBPU_ESCR0
MSR_TC_ESCR0
0
2
1
3C0H
3C2H
3C4H
MSR_MS_COUNTER2
6
306H
MSR_MS_CCCR2
366H
MSR_MS_ESCR1
MSR_TBPU_ESCR1
MSR_TC_ESCR1
0
2
1
3C1H
3C3H
3C5H
MSR_MS_COUNTER3
7
307H
MSR_MS_CCCR3
367H
MSR_MS_ESCR1
MSR_TBPU_ESCR1
MSR_TC_ESCR1
0
2
1
3C1H
3C3H
3C5H
30-66 Vol. 3B
PERFORMANCE MONITORING
Table 30-28. Performance Counter MSRs and Associated CCCR and
ESCR MSRs (Pentium 4 and Intel Xeon Processors) (Contd.)
Counter
CCCR
ESCR
Name
No.
Addr
Name
Addr
Name
No. Addr
MSR_FLAME_
COUNTER0
8
308H
MSR_FLAME_CCCR0
368H
MSR_FIRM_ESCR0
MSR_FLAME_ESCR0
MSR_DAC_ESCR0
MSR_SAAT_ESCR0
MSR_U2L_ESCR0
1
0
5
2
3
3A4H
3A6H
3A8H
3AEH
3B0H
MSR_FLAME_
COUNTER1
9
309H
MSR_FLAME_CCCR1
369H
MSR_FIRM_ESCR0
MSR_FLAME_ESCR0
MSR_DAC_ESCR0
MSR_SAAT_ESCR0
MSR_U2L_ESCR0
1
0
5
2
3
3A4H
3A6H
3A8H
3AEH
3B0H
MSR_FLAME_
COUNTER2
10
30AH
MSR_FLAME_CCCR2
36AH
MSR_FIRM_ESCR1
MSR_FLAME_ESCR1
MSR_DAC_ESCR1
MSR_SAAT_ESCR1
MSR_U2L_ESCR1
1
0
5
2
3
3A5H
3A7H
3A9H
3AFH
3B1H
MSR_FLAME_
COUNTER3
11
30BH
MSR_FLAME_CCCR3
36BH
MSR_FIRM_ESCR1
MSR_FLAME_ESCR1
MSR_DAC_ESCR1
MSR_SAAT_ESCR1
MSR_U2L_ESCR1
1
0
5
2
3
3A5H
3A7H
3A9H
3AFH
3B1H
MSR_IQ_COUNTER0
12
30CH
MSR_IQ_CCCR0
36CH
MSR_CRU_ESCR0
MSR_CRU_ESCR2
MSR_CRU_ESCR4
MSR_IQ_ESCR01
MSR_RAT_ESCR0
MSR_SSU_ESCR0
MSR_ALF_ESCR0
4
5
6
0
2
3
1
3B8H
3CCH
3E0H
3BAH
3BCH
3BEH
3CAH
MSR_IQ_COUNTER1
13
30DH
MSR_IQ_CCCR1
36DH
MSR_CRU_ESCR0
MSR_CRU_ESCR2
MSR_CRU_ESCR4
MSR_IQ_ESCR01
MSR_RAT_ESCR0
MSR_SSU_ESCR0
MSR_ALF_ESCR0
4
5
6
0
2
3
1
3B8H
3CCH
3E0H
3BAH
3BCH
3BEH
3CAH
MSR_IQ_COUNTER2
14
30EH
MSR_IQ_CCCR2
36EH
MSR_CRU_ESCR1
MSR_CRU_ESCR3
MSR_CRU_ESCR5
MSR_IQ_ESCR11
MSR_RAT_ESCR1
MSR_ALF_ESCR1
4
5
6
0
2
1
3B9H
3CDH
3E1H
3BBH
3BDH
3CBH
MSR_IQ_COUNTER3
15
30FH
MSR_IQ_CCCR3
36FH
MSR_CRU_ESCR1
MSR_CRU_ESCR3
MSR_CRU_ESCR5
MSR_IQ_ESCR11
MSR_RAT_ESCR1
MSR_ALF_ESCR1
4
5
6
3B9H
3CDH
3E1H
3BBH
3BDH
3CBH
2
1
0
Vol. 3B 30-67
PERFORMANCE MONITORING
Table 30-28. Performance Counter MSRs and Associated CCCR and
ESCR MSRs (Pentium 4 and Intel Xeon Processors) (Contd.)
Counter
CCCR
ESCR
Name
No.
Addr
Name
Addr
Name
No. Addr
MSR_IQ_COUNTER4
16
310H
MSR_IQ_CCCR4
370H
MSR_CRU_ESCR0
MSR_CRU_ESCR2
MSR_CRU_ESCR4
MSR_IQ_ESCR01
MSR_RAT_ESCR0
MSR_SSU_ESCR0
MSR_ALF_ESCR0
4
5
6
0
2
3
1
3B8H
3CCH
3E0H
3BAH
3BCH
3BEH
3CAH
MSR_IQ_COUNTER5
17
311H
MSR_IQ_CCCR5
371H
MSR_CRU_ESCR1
MSR_CRU_ESCR3
MSR_CRU_ESCR5
MSR_IQ_ESCR11
MSR_RAT_ESCR1
MSR_ALF_ESCR1
4
5
6
0
2
1
3B9H
3CDH
3E1H
3BBH
3BDH
3CBH
NOTES:
1. MSR_IQ_ESCR0 and MSR_IQ_ESCR1 are available only on early processor builds (family 0FH, models 01H-02H). These MSRs are not available on later versions.
The types of events that can be counted with these performance monitoring facilities
are divided into two classes: non-retirement events and at-retirement events.
•
Non-retirement events (see Table A-13) are events that occur any time during
instruction execution (such as bus transactions or cache transactions).
•
At-retirement events (see Table A-14) are events that are counted at the
retirement stage of instruction execution, which allows finer granularity in
counting events and capturing machine state.
The at-retirement counting mechanism includes facilities for tagging μops that
have encountered a particular performance event during instruction execution.
Tagging allows events to be sorted between those that occurred on an execution
path that resulted in architectural state being committed at retirement as well as
events that occurred on an execution path where the results were eventually
cancelled and never committed to architectural state (such as, the execution of a
mispredicted branch).
The Pentium 4 and Intel Xeon processor performance monitoring facilities support
the three usage models described below. The first two models can be used to count
both non-retirement and at-retirement events; the third model is used to count a
subset of at-retirement events:
•
Event counting — A performance counter is configured to count one or more
types of events. While the counter is counting, software reads the counter at
selected intervals to determine the number of events that have been counted
between the intervals.
•
Non-precise event-based sampling — A performance counter is configured to
count one or more types of events and to generate an interrupt when it
30-68 Vol. 3B
PERFORMANCE MONITORING
overflows. To trigger an overflow, the counter is preset to a modulus value that
will cause the counter to overflow after a specific number of events have been
counted.
When the counter overflows, the processor generates a performance monitoring
interrupt (PMI). The interrupt service routine for the PMI then records the return
instruction pointer (RIP), resets the modulus, and restarts the counter. Code
performance can be analyzed by examining the distribution of RIPs with a tool
like the VTune™ Performance Analyzer.
•
Precise event-based sampling (PEBS) — This type of performance
monitoring is similar to non-precise event-based sampling, except that a
memory buffer is used to save a record of the architectural state of the processor
whenever the counter overflows. The records of architectural state provide
additional information for use in performance tuning. Precise event-based
sampling can be used to count only a subset of at-retirement events.
The following sections describe the MSRs and data structures used for performance
monitoring in the Pentium 4 and Intel Xeon processors.
30.9.1
ESCR MSRs
The 45 ESCR MSRs (see Table 30-28) allow software to select specific events to be
countered. Each ESCR is usually associated with a pair of performance counters (see
Table 30-28) and each performance counter has several ESCRs associated with it
(allowing the events counted to be selected from a variety of events).
Figure 30-33 shows the layout of an ESCR MSR. The functions of the flags and fields
are:
•
USR flag, bit 2 — When set, events are counted when the processor is operating
at a current privilege level (CPL) of 1, 2, or 3. These privilege levels are generally
used by application code and unprotected operating system code.
•
OS flag, bit 3 — When set, events are counted when the processor is operating
at CPL of 0. This privilege level is generally reserved for protected operating
system code. (When both the OS and USR flags are set, events are counted at all
privilege levels.)
Vol. 3B 30-69
PERFORMANCE MONITORING
31 30
25 24
Event
Select
5 4 3 2 1 0
9 8
Tag
Value
Event Mask
Tag Enable
OS
USR
Reserved
63
32
Reserved
Figure 30-33. Event Selection Control Register (ESCR) for Pentium 4
and Intel Xeon Processors without Intel HT Technology Support
•
Tag enable, bit 4 — When set, enables tagging of μops to assist in at-retirement
event counting; when clear, disables tagging. See Section 30.9.6, “At-Retirement
Counting.”
•
Tag value field, bits 5 through 8 — Selects a tag value to associate with a μop
to assist in at-retirement event counting.
•
Event mask field, bits 9 through 24 — Selects events to be counted from the
event class selected with the event select field.
•
Event select field, bits 25 through 30) — Selects a class of events to be
counted. The events within this class that are counted are selected with the event
mask field.
When setting up an ESCR, the event select field is used to select a specific class of
events to count, such as retired branches. The event mask field is then used to select
one or more of the specific events within the class to be counted. For example, when
counting retired branches, four different events can be counted: branch not taken
predicted, branch not taken mispredicted, branch taken predicted, and branch taken
mispredicted. The OS and USR flags allow counts to be enabled for events that occur
when operating system code and/or application code are being executed. If neither
the OS nor USR flag is set, no events will be counted.
The ESCRs are initialized to all 0s on reset. The flags and fields of an ESCR are configured by writing to the ESCR using the WRMSR instruction. Table 30-28 gives the
addresses of the ESCR MSRs.
Writing to an ESCR MSR does not enable counting with its associated performance
counter; it only selects the event or events to be counted. The CCCR for the selected
performance counter must also be configured. Configuration of the CCCR includes
selecting the ESCR and enabling the counter.
30-70 Vol. 3B
PERFORMANCE MONITORING
30.9.2
Performance Counters
The performance counters in conjunction with the counter configuration control
registers (CCCRs) are used for filtering and counting the events selected by the
ESCRs. The Pentium 4 and Intel Xeon processors provide 18 performance counters
organized into 9 pairs. A pair of performance counters is associated with a particular
subset of events and ESCR’s (see Table 30-28). The counter pairs are partitioned into
four groups:
•
The BPU group, includes two performance counter pairs:
— MSR_BPU_COUNTER0 and MSR_BPU_COUNTER1.
— MSR_BPU_COUNTER2 and MSR_BPU_COUNTER3.
•
The MS group, includes two performance counter pairs:
— MSR_MS_COUNTER0 and MSR_MS_COUNTER1.
— MSR_MS_COUNTER2 and MSR_MS_COUNTER3.
•
The FLAME group, includes two performance counter pairs:
— MSR_FLAME_COUNTER0 and MSR_FLAME_COUNTER1.
— MSR_FLAME_COUNTER2 and MSR_FLAME_COUNTER3.
•
The IQ group, includes three performance counter pairs:
— MSR_IQ_COUNTER0 and MSR_IQ_COUNTER1.
— MSR_IQ_COUNTER2 and MSR_IQ_COUNTER3.
— MSR_IQ_COUNTER4 and MSR_IQ_COUNTER5.
The MSR_IQ_COUNTER4 counter in the IQ group provides support for the PEBS.
Alternate counters in each group can be cascaded: the first counter in one pair can
start the first counter in the second pair and vice versa. A similar cascading is
possible for the second counters in each pair. For example, within the BPU group of
counters, MSR_BPU_COUNTER0 can start MSR_BPU_COUNTER2 and vice versa, and
MSR_BPU_COUNTER1 can start MSR_BPU_COUNTER3 and vice versa (see Section
30.9.5.6, “Cascading Counters”). The cascade flag in the CCCR register for the
performance counter enables the cascading of counters.
Each performance counter is 40-bits wide (see Figure 30-34). The RDPMC instruction
has been enhanced in the Pentium 4 and Intel Xeon processors to allow reading of
either the full counter-width (40-bits) or the low 32-bits of the counter. Reading the
low 32-bits is faster than reading the full counter width and is appropriate in situations where the count is small enough to be contained in 32 bits.
The RDPMC instruction can be used by programs or procedures running at any privilege level and in virtual-8086 mode to read these counters. The PCE flag in control
register CR4 (bit 8) allows the use of this instruction to be restricted to only programs
and procedures running at privilege level 0.
Vol. 3B 30-71
PERFORMANCE MONITORING
31
0
Counter
63
32
39
Reserved
Counter
Figure 30-34. Performance Counter (Pentium 4 and Intel Xeon Processors)
The RDPMC instruction is not serializing or ordered with other instructions. Thus, it
does not necessarily wait until all previous instructions have been executed before
reading the counter. Similarly, subsequent instructions may begin execution before
the RDPMC instruction operation is performed.
Only the operating system, executing at privilege level 0, can directly manipulate the
performance counters, using the RDMSR and WRMSR instructions. A secure operating system would clear the PCE flag during system initialization to disable direct
user access to the performance-monitoring counters, but provide a user-accessible
programming interface that emulates the RDPMC instruction.
Some uses of the performance counters require the counters to be preset before
counting begins (that is, before the counter is enabled). This can be accomplished by
writing to the counter using the WRMSR instruction. To set a counter to a specified
number of counts before overflow, enter a 2s complement negative integer in the
counter. The counter will then count from the preset value up to -1 and overflow.
Writing to a performance counter in a Pentium 4 or Intel Xeon processor with the
WRMSR instruction causes all 40 bits of the counter to be written.
30.9.3
CCCR MSRs
Each of the 18 performance counters in a Pentium 4 or Intel Xeon processor has one
CCCR MSR associated with it (see Table 30-28). The CCCRs control the filtering and
counting of events as well as interrupt generation. Figure 30-35 shows the layout of
an CCCR MSR. The functions of the flags and fields are as follows:
•
Enable flag, bit 12 — When set, enables counting; when clear, the counter is
disabled. This flag is cleared on reset.
•
ESCR select field, bits 13 through 15 — Identifies the ESCR to be used to
select events to be counted with the counter associated with the CCCR.
•
Compare flag, bit 18 — When set, enables filtering of the event count; when
clear, disables filtering. The filtering method is selected with the threshold,
complement, and edge flags.
•
Complement flag, bit 19 — Selects how the incoming event count is compared
with the threshold value. When set, event counts that are less than or equal to
the threshold value result in a single count being delivered to the performance
30-72 Vol. 3B
PERFORMANCE MONITORING
counter; when clear, counts greater than the threshold value result in a count
being delivered to the performance counter (see Section 30.9.5.2, “Filtering
Events”). The complement flag is not active unless the compare flag is set.
•
Threshold field, bits 20 through 23 — Selects the threshold value to be used
for comparisons. The processor examines this field only when the compare flag is
set, and uses the complement flag setting to determine the type of threshold
comparison to be made. The useful range of values that can be entered in this
field depend on the type of event being counted (see Section 30.9.5.2, “Filtering
Events”).
•
Edge flag, bit 24 — When set, enables rising edge (false-to-true) edge
detection of the threshold comparison output for filtering event counts; when
clear, rising edge detection is disabled. This flag is active only when the compare
flag is set.
Reserved
31 30 29
27 26 25 24 23
20 19 18 17 16 15
Threshold
13 12 11
ESCR
Select
0
Reserved
Reserved
Enable
Reserved: Must be set to 11B
Compare
Complement
Edge
FORCE_OVF
OVF_PMI
Cascade
OVF
63
32
Reserved
Figure 30-35. Counter Configuration Control Register (CCCR)
•
FORCE_OVF flag, bit 25 — When set, forces a counter overflow on every
counter increment; when clear, overflow only occurs when the counter actually
overflows.
•
OVF_PMI flag, bit 26 — When set, causes a performance monitor interrupt
(PMI) to be generated when the counter overflows occurs; when clear, disables
PMI generation. Note that the PMI is generated on the next event count after the
counter has overflowed.
Vol. 3B 30-73
PERFORMANCE MONITORING
•
Cascade flag, bit 30 — When set, enables counting on one counter of a counter
pair when its alternate counter in the other the counter pair in the same counter
group overflows (see Section 30.9.2, “Performance Counters,” for further
details); when clear, disables cascading of counters.
•
OVF flag, bit 31 — Indicates that the counter has overflowed when set. This flag
is a sticky flag that must be explicitly cleared by software.
The CCCRs are initialized to all 0s on reset.
The events that an enabled performance counter actually counts are selected and
filtered by the following flags and fields in the ESCR and CCCR registers and in the
qualification order given:
1. The event select and event mask fields in the ESCR select a class of events to be
counted and one or more event types within the class, respectively.
2. The OS and USR flags in the ESCR selected the privilege levels at which events
will be counted.
3. The ESCR select field of the CCCR selects the ESCR. Since each counter has
several ESCRs associated with it, one ESCR must be chosen to select the classes
of events that may be counted.
4. The compare and complement flags and the threshold field of the CCCR select an
optional threshold to be used in qualifying an event count.
5. The edge flag in the CCCR allows events to be counted only on rising-edge transitions.
The qualification order in the above list implies that the filtered output of one “stage”
forms the input for the next. For instance, events filtered using the privilege level
flags can be further qualified by the compare and complement flags and the
threshold field, and an event that matched the threshold criteria, can be further qualified by edge detection.
The uses of the flags and fields in the CCCRs are discussed in greater detail in Section
30.9.5, “Programming the Performance Counters for Non-Retirement Events.”
30.9.4
Debug Store (DS) Mechanism
The debug store (DS) mechanism was introduced in the Pentium 4 and Intel Xeon
processors to allow various types of information to be collected in memory-resident
buffers for use in debugging and tuning programs. For the Pentium 4 and Intel Xeon
processors, the DS mechanism is used to collect two types of information: branch
records and precise event-based sampling (PEBS) records. The availability of the DS
mechanism in a processor is indicated with the DS feature flag (bit 21) returned by
the CPUID instruction.
See Section 16.4.5, “Branch Trace Store (BTS),” and Section 30.9.7, “Precise EventBased Sampling (PEBS),” for a description of these facilities. Records collected with
the DS mechanism are saved in the DS save area. See Section 16.4.9, “BTS and DS
Save Area.”
30-74 Vol. 3B
PERFORMANCE MONITORING
30.9.5
Programming the Performance Counters
for Non-Retirement Events
The basic steps to program a performance counter and to count events include the
following:
1. Select the event or events to be counted.
2. For each event, select an ESCR that supports the event using the values in the
ESCR restrictions row in Table A-13, Appendix A.
3. Match the CCCR Select value and ESCR name in Table A-13 to a value listed in
Table 30-28; select a CCCR and performance counter.
4. Set up an ESCR for the specific event or events to be counted and the privilege
levels at which the are to be counted.
5. Set up the CCCR for the performance counter by selecting the ESCR and the
desired event filters.
6. Set up the CCCR for optional cascading of event counts, so that when the
selected counter overflows its alternate counter starts.
7. Set up the CCCR to generate an optional performance monitor interrupt (PMI)
when the counter overflows. If PMI generation is enabled, the local APIC must be
set up to deliver the interrupt to the processor and a handler for the interrupt
must be in place.
8. Enable the counter to begin counting.
30.9.5.1
Selecting Events to Count
Table A-14 in Appendix A lists a set of at-retirement events for the Pentium 4 and
Intel Xeon processors. For each event listed in Table A-14, setup information is
provided. Table 30-29 gives an example of one of the events.
Table 30-29. Event Example
Event Name
Event Parameters
Parameter Value
branch_retired
Description
Counts the retirement of a branch.
Specify one or more mask bits to
select any combination of branch
taken, not-taken, predicted and
mispredicted.
ESCR restrictions
MSR_CRU_ESCR2
MSR_CRU_ESCR3
See Table 15-3 for the addresses of
the ESCR MSRs
Counter numbers
per ESCR
ESCR2: 12, 13, 16
The counter numbers associated
with each ESCR are provided. The
performance counters and
corresponding CCCRs can be obtained
from Table 15-3.
ESCR3: 14, 15, 17
Vol. 3B 30-75
PERFORMANCE MONITORING
Table 30-29. Event Example (Contd.)
Event Name
Event Parameters
Parameter Value
Description
ESCR Event Select
06H
ESCR[31:25]
ESCR[24:9],
ESCR Event Mask
Bit 0: MMNP
CCCR Select
Can Support PEBS
1: MMNM
Branch Not-taken Mispredicted,
2: MMTP
Branch Taken Predicted,
3: MMTM
Branch Taken Mispredicted.
05H
Event Specific
Notes
Branch Not-taken Predicted,
CCCR[15:13]
P6: EMON_BR_INST_RETIRED
No
Requires Additional No
MSRs for Tagging
For Table A-13 and Table A-14, Appendix A, the name of the event is listed in the
Event Name column and parameters that define the event and other information are
listed in the Event Parameters column. The Parameter Value and Description columns
give specific parameters for the event and additional description information. Entries
in the Event Parameters column are described below.
•
ESCR restrictions — Lists the ESCRs that can be used to program the event.
Typically only one ESCR is needed to count an event.
•
Counter numbers per ESCR — Lists which performance counters are
associated with each ESCR. Table 30-28 gives the name of the counter and CCCR
for each counter number. Typically only one counter is needed to count the event.
•
ESCR event select — Gives the value to be placed in the event select field of the
ESCR to select the event.
•
ESCR event mask — Gives the value to be placed in the Event Mask field of the
ESCR to select sub-events to be counted. The parameter value column defines
the documented bits with relative bit position offset starting from 0, where the
absolute bit position of relative offset 0 is bit 9 of the ESCR. All undocumented
bits are reserved and should be set to 0.
•
CCCR select — Gives the value to be placed in the ESCR select field of the CCCR
associated with the counter to select the ESCR to be used to define the event.
This value is not the address of the ESCR; it is the number of the ESCR from the
Number column in Table 30-28.
•
Event specific notes — Gives additional information about the event, such as
the name of the same or a similar event defined for the P6 family processors.
•
Can support PEBS — Indicates if PEBS is supported for the event (only supplied
for at-retirement events listed in Table A-14.)
30-76 Vol. 3B
PERFORMANCE MONITORING
•
Requires additional MSR for tagging — Indicates which if any additional
MSRs must be programmed to count the events (only supplied for the atretirement events listed in Table A-14.)
NOTE
The performance-monitoring events listed in Appendix A, “Performance-Monitoring Events,” are intended to be used as guides for
performance tuning. The counter values reported are not guaranteed
to be absolutely accurate and should be used as a relative guide for
tuning. Known discrepancies are documented where applicable.
The following procedure shows how to set up a performance counter for basic
counting; that is, the counter is set up to count a specified event indefinitely, wrapping around whenever it reaches its maximum count. This procedure is continued
through the following four sections.
Using information in Table A-13, Appendix A, an event to be counted can be selected
as follows:
1. Select the event to be counted.
2. Select the ESCR to be used to select events to be counted from the ESCRs field.
3. Select the number of the counter to be used to count the event from the Counter
Numbers Per ESCR field.
4. Determine the name of the counter and the CCCR associated with the counter,
and determine the MSR addresses of the counter, CCCR, and ESCR from Table
30-28.
5. Use the WRMSR instruction to write the ESCR Event Select and ESCR Event Mask
values into the appropriate fields in the ESCR. At the same time set or clear the
USR and OS flags in the ESCR as desired.
6. Use the WRMSR instruction to write the CCCR Select value into the appropriate
field in the CCCR.
NOTE
Typically all the fields and flags of the CCCR will be written with one
WRMSR instruction; however, in this procedure, several WRMSR
writes are used to more clearly demonstrate the uses of the various
CCCR fields and flags.
This setup procedure is continued in the next section, Section 30.9.5.2, “Filtering
Events.”
30.9.5.2
Filtering Events
Each counter receives up to 4 input lines from the processor hardware from which it
is counting events. The counter treats these inputs as binary inputs (input 0 has a
Vol. 3B 30-77
PERFORMANCE MONITORING
value of 1, input 1 has a value of 2, input 3 has a value of 4, and input 3 has a value
of 8). When a counter is enabled, it adds this binary input value to the counter value
on each clock cycle. For each clock cycle, the value added to the counter can then
range from 0 (no event) to 15.
For many events, only the 0 input line is active, so the counter is merely counting the
clock cycles during which the 0 input is asserted. However, for some events two or
more input lines are used. Here, the counters threshold setting can be used to filter
events. The compare, complement, threshold, and edge fields control the filtering of
counter increments by input value.
If the compare flag is set, then a “greater than” or a “less than or equal to” comparison of the input value vs. a threshold value can be made. The complement flag
selects “less than or equal to” (flag set) or “greater than” (flag clear). The threshold
field selects a threshold value of from 0 to 15. For example, if the complement flag is
cleared and the threshold field is set to 6, than any input value of 7 or greater on the
4 inputs to the counter will cause the counter to be incremented by 1, and any value
less than 7 will cause an increment of 0 (or no increment) of the counter. Conversely,
if the complement flag is set, any value from 0 to 6 will increment the counter and
any value from 7 to 15 will not increment the counter. Note that when a threshold
condition has been satisfied, the input to the counter is always 1, not the input value
that is presented to the threshold filter.
The edge flag provides further filtering of the counter inputs when a threshold
comparison is being made. The edge flag is only active when the compare flag is set.
When the edge flag is set, the resulting output from the threshold filter (a value of 0
or 1) is used as an input to the edge filter. Each clock cycle, the edge filter examines
the last and current input values and sends a count to the counter only when it
detects a “rising edge” event; that is, a false-to-true transition. Figure 30-36 illustrates rising edge filtering.
The following procedure shows how to configure a CCCR to filter events using the
threshold filter and the edge filter. This procedure is a continuation of the setup
procedure introduced in Section 30.9.5.1, “Selecting Events to Count.”
7. (Optional) To set up the counter for threshold filtering, use the WRMSR
instruction to write values in the CCCR compare and complement flags and the
threshold field:
— Set the compare flag.
— Set or clear the complement flag for less than or equal to or greater than
comparisons, respectively.
— Enter a value from 0 to 15 in the threshold field.
8. (Optional) Select rising edge filtering by setting the CCCR edge flag.
This setup procedure is continued in the next section, Section 30.9.5.3, “Starting
Event Counting.”
30-78 Vol. 3B
PERFORMANCE MONITORING
Processor Clock
Output from
Threshold Filter
Counter Increments
On Rising Edge
(False-to-True)
Figure 30-36. Effects of Edge Filtering
30.9.5.3
Starting Event Counting
Event counting by a performance counter can be initiated in either of two ways. The
typical way is to set the enable flag in the counter’s CCCR. Following the instruction
to set the enable flag, event counting begins and continues until it is stopped (see
Section 30.9.5.5, “Halting Event Counting”).
The following procedural step shows how to start event counting. This step is a
continuation of the setup procedure introduced in Section 30.9.5.2, “Filtering
Events.”
9. To start event counting, use the WRMSR instruction to set the CCCR enable flag
for the performance counter.
This setup procedure is continued in the next section, Section 30.9.5.4, “Reading a
Performance Counter’s Count.”
The second way that a counter can be started by using the cascade feature. Here, the
overflow of one counter automatically starts its alternate counter (see Section
30.9.5.6, “Cascading Counters”).
30.9.5.4
Reading a Performance Counter’s Count
The Pentium 4 and Intel Xeon processors’ performance counters can be read using
either the RDPMC or RDMSR instructions. The enhanced functions of the RDPMC
instruction (including fast read) are described in Section 30.9.2, “Performance
Counters.” These instructions can be used to read a performance counter while it is
counting or when it is stopped.
The following procedural step shows how to read the event counter. This step is a
continuation of the setup procedure introduced in Section 30.9.5.3, “Starting Event
Counting.”
10. To read a performance counters current event count, execute the RDPMC
instruction with the counter number obtained from Table 30-28 used as an
operand.
Vol. 3B 30-79
PERFORMANCE MONITORING
This setup procedure is continued in the next section, Section 30.9.5.5, “Halting
Event Counting.”
30.9.5.5
Halting Event Counting
After a performance counter has been started (enabled), it continues counting indefinitely. If the counter overflows (goes one count past its maximum count), it wraps
around and continues counting. When the counter wraps around, it sets its OVF flag
to indicate that the counter has overflowed. The OVF flag is a sticky flag that indicates that the counter has overflowed at least once since the OVF bit was last
cleared.
To halt counting, the CCCR enable flag for the counter must be cleared.
The following procedural step shows how to stop event counting. This step is a
continuation of the setup procedure introduced in Section 30.9.5.4, “Reading a
Performance Counter’s Count.”
11. To stop event counting, execute a WRMSR instruction to clear the CCCR enable
flag for the performance counter.
To halt a cascaded counter (a counter that was started when its alternate counter
overflowed), either clear the Cascade flag in the cascaded counter’s CCCR MSR or
clear the OVF flag in the alternate counter’s CCCR MSR.
30.9.5.6
Cascading Counters
As described in Section 30.9.2, “Performance Counters,” eighteen performance
counters are implemented in pairs. Nine pairs of counters and associated CCCRs are
further organized as four blocks: BPU, MS, FLAME, and IQ (see Table 30-28). The first
three blocks contain two pairs each. The IQ block contains three pairs of counters (12
through 17) with associated CCCRs (MSR_IQ_CCCR0 through MSR_IQ_CCCR5).
The first 8 counter pairs (0 through 15) can be programmed using ESCRs to detect
performance monitoring events. Pairs of ESCRs in each of the four blocks allow many
different types of events to be counted. The cascade flag in the CCCR MSR allows
nested monitoring of events to be performed by cascading one counter to a second
counter located in another pair in the same block (see Figure 30-35 for the location
of the flag).
Counters 0 and 1 form the first pair in the BPU block. Either counter 0 or 1 can be
programmed to detect an event via MSR_MO B_ESCR0. Counters 0 and 2 can be
cascaded in any order, as can counters 1 and 3. It’s possible to set up 4 counters in
the same block to cascade on two pairs of independent events. The pairing described
also applies to subsequent blocks. Since the IQ PUB has two extra counters,
cascading operates somewhat differently if 16 and 17 are involved. In the IQ block,
counter 16 can only be cascaded from counter 14 (not from 12); counter 14 cannot
be cascaded from counter 16 using the CCCR cascade bit mechanism. Similar restrictions apply to counter 17.
30-80 Vol. 3B
PERFORMANCE MONITORING
Example 30-1. Counting Events
Assume a scenario where counter X is set up to count 200 occurrences of event A;
then counter Y is set up to count 400 occurrences of event B. Each counter is set up
to count a specific event and overflow to the next counter. In the above example,
counter X is preset for a count of -200 and counter Y for a count of -400; this setup
causes the counters to overflow on the 200th and 400th counts respectively.
Continuing this scenario, counter X is set up to count indefinitely and wraparound on
overflow. This is described in the basic performance counter setup procedure that
begins in Section 30.9.5.1, “Selecting Events to Count.” Counter Y is set up with the
cascade flag in its associated CCCR MSR set to 1 and its enable flag set to 0.
To begin the nested counting, the enable bit for the counter X is set. Once enabled,
counter X counts until it overflows. At this point, counter Y is automatically enabled
and begins counting. Thus counter X overflows after 200 occurrences of event A.
Counter Y then starts, counting 400 occurrences of event B before overflowing. When
performance counters are cascaded, the counter Y would typically be set up to
generate an interrupt on overflow. This is described in Section 30.9.5.8, “Generating
an Interrupt on Overflow.”
The cascading counters mechanism can be used to count a single event. The
counting begins on one counter then continues on the second counter after the first
counter overflows. This technique doubles the number of event counts that can be
recorded, since the contents of the two counters can be added together.
30.9.5.7
EXTENDED CASCADING
Extended cascading is a model-specific feature in the Intel NetBurst microarchitecture. The feature is available to Pentium 4 and Xeon processor family with family
encoding of 15 and model encoding greater than or equal to 2. This feature uses bit
11 in CCCRs associated with the IQ block. See Table 30-30.
Table 30-30. CCR Names and Bit Positions
CCCR Name:Bit Position
Bit Name
Description
MSR_IQ_CCCR1|2:11
Reserved
MSR_IQ_CCCR0:11
CASCNT4INTO0
Allow counter 4 to cascade into
counter 0
MSR_IQ_CCCR3:11
CASCNT5INTO3
Allow counter 5 to cascade into
counter 3
MSR_IQ_CCCR4:11
CASCNT5INTO4
Allow counter 5 to cascade into
counter 4
MSR_IQ_CCCR5:11
CASCNT4INTO5
Allow counter 4 to cascade into
counter 5
Vol. 3B 30-81
PERFORMANCE MONITORING
The extended cascading feature can be adapted to the sampling usage model for
performance monitoring. However, it is known that performance counters do not
generate PMI in cascade mode or extended cascade mode due to an erratum. This
erratum applies to Pentium 4 and Intel Xeon processors with model encoding of 2.
For Pentium 4 and Intel Xeon processors with model encoding of 0 and 1, the erratum
applies to processors with stepping encoding greater than 09H.
Counters 16 and 17 in the IQ block are frequently used in precise event-based
sampling or at-retirement counting of events indicating a stalled condition in the
pipeline. Neither counter 16 or 17 can initiate the cascading of counter pairs using
the cascade bit in a CCCR.
Extended cascading permits performance monitoring tools to use counters 16 and 17
to initiate cascading of two counters in the IQ block. Extended cascading from
counter 16 and 17 is conceptually similar to cascading other counters, but instead of
using CASCADE bit of a CCCR, one of the four CASCNTxINTOy bits is used.
Example 30-2. Scenario for Extended Cascading
A usage scenario for extended cascading is to sample instructions retired on logical
processor 1 after the first 4096 instructions retired on logical processor 0. A procedure to program extended cascading in this scenario is outlined below:
1. Write the value 0 to counter 12.
2. Write the value 04000603H to MSR_CRU_ESCR0 (corresponding to selecting the
NBOGNTAG and NBOGTAG event masks with qualification restricted to logical
processor 1).
3. Write the value 04038800H to MSR_IQ_CCCR0. This enables CASCNT4INTO0
and OVF_PMI. An ISR can sample on instruction addresses in this case (do not
set ENABLE, or CASCADE).
4. Write the value FFFFF000H into counter 16.1.
5. Write the value 0400060CH to MSR_CRU_ESCR2 (corresponding to selecting the
NBOGNTAG and NBOGTAG event masks with qualification restricted to logical
processor 0).
6. Write the value 00039000H to MSR_IQ_CCCR4 (set ENABLE bit, but not
OVF_PMI).
Another use for cascading is to locate stalled execution in a multithreaded application. Assume MOB replays in thread B cause thread A to stall. Getting a sample of the
stalled execution in this scenario could be accomplished by:
1. Set up counter B to count MOB replays on thread B.
2. Set up counter A to count resource stalls on thread A; set its force overflow bit
and the appropriate CASCNTxINTOy bit.
3. Use the performance monitoring interrupt to capture the program execution data
of the stalled thread.
30-82 Vol. 3B
PERFORMANCE MONITORING
30.9.5.8
Generating an Interrupt on Overflow
Any performance counter can be configured to generate a performance monitor
interrupt (PMI) if the counter overflows. The PMI interrupt service routine can then
collect information about the state of the processor or program when overflow
occurred. This information can then be used with a tool like the Intel® VTune™
Performance Analyzer to analyze and tune program performance.
To enable an interrupt on counter overflow, the OVR_PMI flag in the counter’s associated CCCR MSR must be set. When overflow occurs, a PMI is generated through the
local APIC. (Here, the performance counter entry in the local vector table [LVT] is set
up to deliver the interrupt generated by the PMI to the processor.)
The PMI service routine can use the OVF flag to determine which counter overflowed
when multiple counters have been configured to generate PMIs. Also, note that these
processors mask PMIs upon receiving an interrupt. Clear this condition before leaving
the interrupt handler.
When generating interrupts on overflow, the performance counter being used should
be preset to value that will cause an overflow after a specified number of events are
counted plus 1. The simplest way to select the preset value is to write a negative
number into the counter, as described in Section 30.9.5.6, “Cascading Counters.”
Here, however, if an interrupt is to be generated after 100 event counts, the counter
should be preset to minus 100 plus 1 (-100 + 1), or -99. The counter will then overflow after it counts 99 events and generate an interrupt on the next (100th) event
counted. The difference of 1 for this count enables the interrupt to be generated
immediately after the selected event count has been reached, instead of waiting for
the overflow to be propagation through the counter.
Because of latency in the microarchitecture between the generation of events and
the generation of interrupts on overflow, it is sometimes difficult to generate an
interrupt close to an event that caused it. In these situations, the FORCE_OVF flag in
the CCCR can be used to improve reporting. Setting this flag causes the counter to
overflow on every counter increment, which in turn triggers an interrupt after every
counter increment.
30.9.5.9
Counter Usage Guideline
There are some instances where the user must take care to configure counting logic
properly, so that it is not powered down. To use any ESCR, even when it is being used
just for tagging, (any) one of the counters that the particular ESCR (or its paired
ESCR) can be connected to should be enabled. If this is not done, 0 counts may
result. Likewise, to use any counter, there must be some event selected in a corresponding ESCR (other than no_event, which generally has a select value of 0).
Vol. 3B 30-83
PERFORMANCE MONITORING
30.9.6
At-Retirement Counting
At-retirement counting provides a means counting only events that represent work
committed to architectural state and ignoring work that was performed speculatively
and later discarded.
The Intel NetBurst microarchitecture used in the Pentium 4 and Intel Xeon processors performs many speculative activities in an attempt to increase effective
processing speeds. One example of this speculative activity is branch prediction. The
Pentium 4 and Intel Xeon processors typically predict the direction of branches and
then decode and execute instructions down the predicted path in anticipation of the
actual branch decision. When a branch misprediction occurs, the results of instructions that were decoded and executed down the mispredicted path are canceled. If a
performance counter was set up to count all executed instructions, the count would
include instructions whose results were canceled as well as those whose results
committed to architectural state.
To provide finer granularity in event counting in these situations, the performance
monitoring facilities provided in the Pentium 4 and Intel Xeon processors provide a
mechanism for tagging events and then counting only those tagged events that
represent committed results. This mechanism is called “at-retirement counting.”
Tables A-14 through A-18 list predefined at-retirement events and event metrics that
can be used to for tagging events when using at retirement counting. The following
terminology is used in describing at-retirement counting:
•
•
Bogus, non-bogus, retire — In at-retirement event descriptions, the term
“bogus” refers to instructions or μops that must be canceled because they are on
a path taken from a mispredicted branch. The terms “retired” and “non-bogus”
refer to instructions or μops along the path that results in committed architectural state changes as required by the program being executed. Thus instructions
and μops are either bogus or non-bogus, but not both. Several of the Pentium 4
and Intel Xeon processors’ performance monitoring events (such as,
Instruction_Retired and Uops_Retired in Table A-14) can count instructions or
μops that are retired based on the characterization of bogus” versus non-bogus.
Tagging — Tagging is a means of marking μops that have encountered a
particular performance event so they can be counted at retirement. During the
course of execution, the same event can happen more than once per μop and a
direct count of the event would not provide an indication of how many μops
encountered that event.
The tagging mechanisms allow a μop to be tagged once during its lifetime and
thus counted once at retirement. The retired suffix is used for performance
metrics that increment a count once per μop, rather than once per event. For
example, a μop may encounter a cache miss more than once during its life time,
but a “Miss Retired” metric (that counts the number of retired μops that
encountered a cache miss) will increment only once for that μop. A “Miss Retired”
metric would be useful for characterizing the performance of the cache hierarchy
for a particular instruction sequence. Details of various performance metrics and
how these can be constructed using the Pentium 4 and Intel Xeon processors
30-84 Vol. 3B
PERFORMANCE MONITORING
performance events are provided in the Intel Pentium 4 Processor Optimization
Reference Manual (see Section 1.4, “Related Literature”).
•
Replay — To maximize performance for the common case, the Intel NetBurst
microarchitecture aggressively schedules μops for execution before all the
conditions for correct execution are guaranteed to be satisfied. In the event that
all of these conditions are not satisfied, μops must be reissued. The mechanism
that the Pentium 4 and Intel Xeon processors use for this reissuing of μops is
called replay. Some examples of replay causes are cache misses, dependence
violations, and unforeseen resource constraints. In normal operation, some
number of replays is common and unavoidable. An excessive number of replays
is an indication of a performance problem.
•
Assist — When the hardware needs the assistance of microcode to deal with
some event, the machine takes an assist. One example of this is an underflow
condition in the input operands of a floating-point operation. The hardware must
internally modify the format of the operands in order to perform the computation.
Assists clear the entire machine of μops before they begin and are costly.
30.9.6.1
Using At-Retirement Counting
The Pentium 4 and Intel Xeon processors allow counting both events and μops that
encountered a specified event. For a subset of the at-retirement events listed in Table
A-14, a μop may be tagged when it encounters that event. The tagging mechanisms
can be used in non-precise event-based sampling, and a subset of these mechanisms
can be used in PEBS. There are four independent tagging mechanisms, and each
mechanism uses a different event to count μops tagged with that mechanism:
•
Front-end tagging — This mechanism pertains to the tagging of μops that
encountered front-end events (for example, trace cache and instruction counts)
and are counted with the Front_end_event event
•
Execution tagging — This mechanism pertains to the tagging of μops that
encountered execution events (for example, instruction types) and are counted
with the Execution_Event event.
•
Replay tagging — This mechanism pertains to tagging of μops whose
retirement is replayed (for example, a cache miss) and are counted with the
Replay_event event. Branch mispredictions are also tagged with this mechanism.
•
No tags — This mechanism does not use tags. It uses the Instr_retired and the
Uops_ retired events.
Each tagging mechanism is independent from all others; that is, a μop that has been
tagged using one mechanism will not be detected with another mechanism’s taggedμop detector. For example, if μops are tagged using the front-end tagging mechanisms, the Replay_event will not count those as tagged μops unless they are also
tagged using the replay tagging mechanism. However, execution tags allow up to
four different types of μops to be counted at retirement through execution tagging.
The independence of tagging mechanisms does not hold when using PEBS. When
using PEBS, only one tagging mechanism should be used at a time.
Vol. 3B 30-85
PERFORMANCE MONITORING
Certain kinds of μops that cannot be tagged, including I/O, uncacheable and locked
accesses, returns, and far transfers.
Table A-14 lists the performance monitoring events that support at-retirement
counting: specifically the Front_end_event, Execution_event, Replay_event,
Inst_retired and Uops_retired events. The following sections describe the tagging
mechanisms for using these events to tag μop and count tagged μops.
30.9.6.2
Tagging Mechanism for Front_end_event
The Front_end_event counts μops that have been tagged as encountering any of the
following events:
•
μop decode events — Tagging μops for μop decode events requires specifying
bits in the ESCR associated with the performance-monitoring event, Uop_type.
•
Trace cache events — Tagging μops for trace cache events may require
specifying certain bits in the MSR_TC_PRECISE_EVENT MSR (see Table A-16).
Table A-14 describes the Front_end_event and Table A-16 describes metrics that are
used to set up a Front_end_event count.
The MSRs specified in the Table A-14 that are supported by the front-end tagging
mechanism must be set and one or both of the NBOGUS and BOGUS bits in the
Front_end_event event mask must be set to count events. None of the events
currently supported requires the use of the MSR_TC_PRECISE_EVENT MSR.
30.9.6.3
Tagging Mechanism For Execution_event
Table A-14 describes the Execution_event and Table A-17 describes metrics that are
used to set up an Execution_event count.
The execution tagging mechanism differs from other tagging mechanisms in how it
causes tagging. One upstream ESCR is used to specify an event to detect and to
specify a tag value (bits 5 through 8) to identify that event. A second downstream
ESCR is used to detect μops that have been tagged with that tag value identifier using
Execution_event for the event selection.
The upstream ESCR that counts the event must have its tag enable flag (bit 4) set
and must have an appropriate tag value mask entered in its tag value field. The 4-bit
tag value mask specifies which of tag bits should be set for a particular μop. The
value selected for the tag value should coincide with the event mask selected in the
downstream ESCR. For example, if a tag value of 1 is set, then the event mask of
NBOGUS0 should be enabled, correspondingly in the downstream ESCR. The downstream ESCR detects and counts tagged μops. The normal (not tag value) mask bits
in the downstream ESCR specify which tag bits to count. If any one of the tag bits
selected by the mask is set, the related counter is incremented by one. This mechanism is summarized in the Table A-17 metrics that are supported by the execution
tagging mechanism. The tag enable and tag value bits are irrelevant for the downstream ESCR used to select the Execution_event.
30-86 Vol. 3B
PERFORMANCE MONITORING
The four separate tag bits allow the user to simultaneously but distinctly count up to
four execution events at retirement. (This applies for non-precise event-based
sampling. There are additional restrictions for PEBS as noted in Section 30.9.7.3,
“Setting Up the PEBS Buffer.”) It is also possible to detect or count combinations of
events by setting multiple tag value bits in the upstream ESCR or multiple mask bits
in the downstream ESCR. For example, use a tag value of 3H in the upstream ESCR
and use NBOGUS0/NBOGUS1 in the downstream ESCR event mask.
30.9.6.4
Tagging Mechanism for Replay_event
Table A-14 describes the Replay_event and Table A-18 describes metrics that are
used to set up an Replay_event count.
The replay mechanism enables tagging of μops for a subset of all replays before
retirement. Use of the replay mechanism requires selecting the type of μop that may
experience the replay in the MSR_PEBS_MATRIX_VERT MSR and selecting the type of
event in the MSR_PEBS_ENABLE MSR. Replay tagging must also be enabled with the
UOP_Tag flag (bit 24) in the MSR_PEBS_ENABLE MSR.
The Table A-18 lists the metrics that are support the replay tagging mechanism and
the at-retirement events that use the replay tagging mechanism, and specifies how
the appropriate MSRs need to be configured. The replay tags defined in Table A-5
also enable Precise Event-Based Sampling (PEBS, see Section 15.9.8). Each of these
replay tags can also be used in normal sampling by not setting Bit 24 nor Bit 25 in
IA_32_PEBS_ENABLE_MSR. Each of these metrics requires that the Replay_Event
(see Table A-14) be used to count the tagged μops.
30.9.7
Precise Event-Based Sampling (PEBS)
The debug store (DS) mechanism in processors based on Intel NetBurst microarchitecture allow two types of information to be collected for use in debugging and tuning
programs: PEBS records and BTS records. See Section 16.4.5, “Branch Trace Store
(BTS),” for a description of the BTS mechanism.
PEBS permits the saving of precise architectural information associated with one or
more performance events in the precise event records buffer, which is part of the DS
save area (see Section 16.4.9, “BTS and DS Save Area”). To use this mechanism, a
counter is configured to overflow after it has counted a preset number of events.
After the counter overflows, the processor copies the current state of the generalpurpose and EFLAGS registers and instruction pointer into a record in the precise
event records buffer. The processor then resets the count in the performance counter
and restarts the counter. When the precise event records buffer is nearly full, an
interrupt is generated, allowing the precise event records to be saved. A circular
buffer is not supported for precise event records.
PEBS is supported only for a subset of the at-retirement events: Execution_event,
Front_end_event, and Replay_event. Also, PEBS can only be carried out using the
one performance counter, the MSR_IQ_COUNTER4 MSR.
Vol. 3B 30-87
PERFORMANCE MONITORING
In processors based on Intel Core microarchitecture, a similar PEBS mechanism is
also supported using IA32_PMC0 and IA32_PERFEVTSEL0 MSRs (See Section
30.4.4).
30.9.7.1
Detection of the Availability of the PEBS Facilities
The DS feature flag (bit 21) returned by the CPUID instruction indicates (when set)
the availability of the DS mechanism in the processor, which supports the PEBS (and
BTS) facilities. When this bit is set, the following PEBS facilities are available:
•
The PEBS_UNAVAILABLE flag in the IA32_MISC_ENABLE MSR indicates (when
clear) the availability of the PEBS facilities, including the MSR_PEBS_ENABLE
MSR.
•
The enable PEBS flag (bit 24) in the MSR_PEBS_ENABLE MSR allows PEBS to be
enabled (set) or disabled (clear).
•
The IA32_DS_AREA MSR can be programmed to point to the DS save area.
30.9.7.2
Setting Up the DS Save Area
Section 16.4.9.2, “Setting Up the DS Save Area,” describes how to set up and enable
the DS save area. This procedure is common for PEBS and BTS.
30.9.7.3
Setting Up the PEBS Buffer
Only the MSR_IQ_COUNTER4 performance counter can be used for PEBS. Use the
following procedure to set up the processor and this counter for PEBS:
1. Set up the precise event buffering facilities. Place values in the precise event
buffer base, precise event index, precise event absolute maximum, and precise
event interrupt threshold, and precise event counter reset fields of the DS buffer
management area (see Figure 16-5) to set up the precise event records buffer in
memory.
2. Enable PEBS. Set the Enable PEBS flag (bit 24) in MSR_PEBS_ENABLE MSR.
3. Set up the MSR_IQ_COUNTER4 performance counter and its associated CCCR
and one or more ESCRs for PEBS as described in Tables A-14 through A-18.
30.9.7.4
Writing a PEBS Interrupt Service Routine
The PEBS facilities share the same interrupt vector and interrupt service routine
(called the DS ISR) with the non-precise event-based sampling and BTS facilities. To
handle PEBS interrupts, PEBS handler code must be included in the DS ISR. See
Section 16.4.9.5, “Writing the DS Interrupt Service Routine,” for guidelines for
writing the DS ISR.
30-88 Vol. 3B
PERFORMANCE MONITORING
30.9.7.5
Other DS Mechanism Implications
The DS mechanism is not available in the SMM. It is disabled on transition to the SMM
mode. Similarly the DS mechanism is disabled on the generation of a machine check
exception and is cleared on processor RESET and INIT.
The DS mechanism is available in real address mode.
30.9.8
Operating System Implications
The DS mechanism can be used by the operating system as a debugging extension to
facilitate failure analysis. When using this facility, a 25 to 30 times slowdown can be
expected due to the effects of the trace store occurring on every taken branch.
Depending upon intended usage, the instruction pointers that are part of the branch
records or the PEBS records need to have an association with the corresponding
process. One solution requires the ability for the DS specific operating system
module to be chained to the context switch. A separate buffer can then be maintained for each process of interest and the MSR pointing to the configuration area
saved and setup appropriately on each context switch.
If the BTS facility has been enabled, then it must be disabled and state stored on
transition of the system to a sleep state in which processor context is lost. The state
must be restored on return from the sleep state.
It is required that an interrupt gate be used for the DS interrupt as opposed to a trap
gate to prevent the generation of an endless interrupt loop.
Pages that contain buffers must have mappings to the same physical address for all
processes/logical processors, such that any change to CR3 will not change DS
addresses. If this requirement cannot be satisfied (that is, the feature is enabled on
a per thread/process basis), then the operating system must ensure that the feature
is enabled/disabled appropriately in the context switch code.
30.10
PERFORMANCE MONITORING AND INTEL HYPERTHREADING TECHNOLOGY IN PROCESSORS BASED
ON INTEL NETBURST® MICROARCHITECTURE
The performance monitoring capability of processors based on Intel NetBurst
microarchitecture and supporting Intel Hyper-Threading Technology is similar to that
described in Section 30.9. However, the capability is extended so that:
•
Performance counters can be programmed to select events qualified by logical
processor IDs.
•
Performance monitoring interrupts can be directed to a specific logical processor
within the physical processor.
Vol. 3B 30-89
PERFORMANCE MONITORING
The sections below describe performance counters, event qualification by logical
processor ID, and special purpose bits in ESCRs/CCCRs. They also describe
MSR_PEBS_ENABLE, MSR_PEBS_MATRIX_VERT, and MSR_TC_PRECISE_EVENT.
30.10.1 ESCR MSRs
Figure 30-37 shows the layout of an ESCR MSR in processors supporting Intel HyperThreading Technology.
The functions of the flags and fields are as follows:
•
T1_USR flag, bit 0 — When set, events are counted when thread 1 (logical
processor 1) is executing at a current privilege level (CPL) of 1, 2, or 3. These
privilege levels are generally used by application code and unprotected operating
system code.
Reserved
31 30
25 24
Event
Select
5 4 3 2 1 0
9 8
Tag
Value
Event Mask
Tag Enable
T0_OS
T0_USR
T1_OS
T1_USR
63
32
Reserved
Figure 30-37. Event Selection Control Register (ESCR) for the Pentium 4 Processor,
Intel Xeon Processor and Intel Xeon Processor MP Supporting Hyper-Threading
Technology
•
T1_OS flag, bit 1 — When set, events are counted when thread 1 (logical
processor 1) is executing at CPL of 0. This privilege level is generally reserved for
protected operating system code. (When both the T1_OS and T1_USR flags are
set, thread 1 events are counted at all privilege levels.)
•
T0_USR flag, bit 2 — When set, events are counted when thread 0 (logical
processor 0) is executing at a CPL of 1, 2, or 3.
•
T0_OS flag, bit 3 — When set, events are counted when thread 0 (logical
processor 0) is executing at CPL of 0. (When both the T0_OS and T0_USR flags
are set, thread 0 events are counted at all privilege levels.)
30-90 Vol. 3B
PERFORMANCE MONITORING
•
Tag enable, bit 4 — When set, enables tagging of μops to assist in at-retirement
event counting; when clear, disables tagging. See Section 30.9.6, “At-Retirement
Counting.”
•
Tag value field, bits 5 through 8 — Selects a tag value to associate with a μop
to assist in at-retirement event counting.
•
Event mask field, bits 9 through 24 — Selects events to be counted from the
event class selected with the event select field.
•
Event select field, bits 25 through 30) — Selects a class of events to be
counted. The events within this class that are counted are selected with the event
mask field.
The T0_OS and T0_USR flags and the T1_OS and T1_USR flags allow event counting
and sampling to be specified for a specific logical processor (0 or 1) within an Intel
Xeon processor MP (See also: Section 8.4.5, “Identifying Logical Processors in an MP
System,” in the Intel® 64 and IA-32 Architectures Software Developer’s Manual,
Volume 3A).
Not all performance monitoring events can be detected within an Intel Xeon
processor MP on a per logical processor basis (see Section 30.10.4, “Performance
Monitoring Events”). Some sub-events (specified by an event mask bits) are counted
or sampled without regard to which logical processor is associated with the detected
event.
30.10.2 CCCR MSRs
Figure 30-38 shows the layout of a CCCR MSR in processors supporting Intel HyperThreading Technology. The functions of the flags and fields are as follows:
•
Enable flag, bit 12 — When set, enables counting; when clear, the counter is
disabled. This flag is cleared on reset
•
ESCR select field, bits 13 through 15 — Identifies the ESCR to be used to
select events to be counted with the counter associated with the CCCR.
•
Active thread field, bits 16 and 17 — Enables counting depending on which
logical processors are active (executing a thread). This field enables filtering of
events based on the state (active or inactive) of the logical processors. The
encodings of this field are as follows:
00 — None. Count only when neither logical processor is active.
01 — Single. Count only when one logical processor is active (either 0 or 1).
10 — Both. Count only when both logical processors are active.
11 — Any. Count when either logical processor is active.
A halted logical processor or a logical processor in the “wait for SIPI” state is
considered inactive.
Vol. 3B 30-91
PERFORMANCE MONITORING
•
Compare flag, bit 18 — When set, enables filtering of the event count; when
clear, disables filtering. The filtering method is selected with the threshold,
complement, and edge flags.
Reserved
31 30 29
27 26 25 24 23
20 19 18 17 16 15
Threshold
13 12 11
ESCR
Select
0
Reserved
Reserved
Enable
Active Thread
Compare
Complement
Edge
FORCE_OVF
OVF_PMI_T0
OVF_PMI_T1
Cascade
OVF
63
32
Reserved
Figure 30-38. Counter Configuration Control Register (CCCR)
•
Complement flag, bit 19 — Selects how the incoming event count is compared
with the threshold value. When set, event counts that are less than or equal to
the threshold value result in a single count being delivered to the performance
counter; when clear, counts greater than the threshold value result in a count
being delivered to the performance counter (see Section 30.9.5.2, “Filtering
Events”). The compare flag is not active unless the compare flag is set.
•
Threshold field, bits 20 through 23 — Selects the threshold value to be used
for comparisons. The processor examines this field only when the compare flag is
set, and uses the complement flag setting to determine the type of threshold
comparison to be made. The useful range of values that can be entered in this
field depend on the type of event being counted (see Section 30.9.5.2, “Filtering
Events”).
•
Edge flag, bit 24 — When set, enables rising edge (false-to-true) edge
detection of the threshold comparison output for filtering event counts; when
clear, rising edge detection is disabled. This flag is active only when the compare
flag is set.
30-92 Vol. 3B
PERFORMANCE MONITORING
•
FORCE_OVF flag, bit 25 — When set, forces a counter overflow on every
counter increment; when clear, overflow only occurs when the counter actually
overflows.
•
OVF_PMI_T0 flag, bit 26 — When set, causes a performance monitor interrupt
(PMI) to be sent to logical processor 0 when the counter overflows occurs; when
clear, disables PMI generation for logical processor 0. Note that the PMI is
generate on the next event count after the counter has overflowed.
•
OVF_PMI_T1 flag, bit 27 — When set, causes a performance monitor interrupt
(PMI) to be sent to logical processor 1 when the counter overflows occurs; when
clear, disables PMI generation for logical processor 1. Note that the PMI is
generate on the next event count after the counter has overflowed.
•
Cascade flag, bit 30 — When set, enables counting on one counter of a counter
pair when its alternate counter in the other the counter pair in the same counter
group overflows (see Section 30.9.2, “Performance Counters,” for further
details); when clear, disables cascading of counters.
•
OVF flag, bit 31 — Indicates that the counter has overflowed when set. This flag
is a sticky flag that must be explicitly cleared by software.
30.10.3 IA32_PEBS_ENABLE MSR
In a processor supporting Intel Hyper-Threading Technology and based on the Intel
NetBurst microarchitecture, PEBS is enabled and qualified with two bits in the
MSR_PEBS_ENABLE MSR: bit 25 (ENABLE_PEBS_MY_THR) and 26
(ENABLE_PEBS_OTH_THR) respectively. These bits do not explicitly identify a
specific logical processor by logic processor ID(T0 or T1); instead, they allow a software agent to enable PEBS for subsequent threads of execution on the same logical
processor on which the agent is running (“my thread”) or for the other logical
processor in the physical package on which the agent is not running (“other thread”).
PEBS is supported for only a subset of the at-retirement events: Execution_event,
Front_end_event, and Replay_event. Also, PEBS can be carried out only with two
performance counters: MSR_IQ_CCCR4 (MSR address 370H) for logical processor 0
and MSR_IQ_CCCR5 (MSR address 371H) for logical processor 1.
Performance monitoring tools should use a processor affinity mask to bind the kernel
mode components that need to modify the ENABLE_PEBS_MY_THR and
ENABLE_PEBS_OTH_THR bits in the MSR_PEBS_ENABLE MSR to a specific logical
processor. This is to prevent these kernel mode components from migrating between
different logical processors due to OS scheduling.
30.10.4 Performance Monitoring Events
All of the events listed in Table A-13 and A-14 are available in an Intel Xeon processor
MP. When Intel Hyper-Threading Technology is active, many performance monitoring
events can be can be qualified by the logical processor ID, which corresponds to bit 0
Vol. 3B 30-93
PERFORMANCE MONITORING
of the initial APIC ID. This allows for counting an event in any or all of the logical
processors. However, not all the events have this logic processor specificity, or thread
specificity.
Here, each event falls into one of two categories:
•
Thread specific (TS) — The event can be qualified as occurring on a specific
logical processor.
•
Thread independent (TI) — The event cannot be qualified as being associated
with a specific logical processor.
Table A-19 gives logical processor specific information (TS or TI) for each of the
events described in Tables A-13 and A-14. If for example, a TS event occurred in
logical processor T0, the counting of the event (as shown in Table 30-31) depends
only on the setting of the T0_USR and T0_OS flags in the ESCR being used to set up
the event counter. The T1_USR and T1_OS flags have no effect on the count.
Table 30-31. Effect of Logical Processor and CPL Qualification
for Logical-Processor-Specific (TS) Events
T1_OS/T1_USR =
00
T1_OS/T1_USR =
01
T1_OS/T1_USR =
11
T1_OS/T1_USR =
10
T0_OS/T0_USR
= 00
Zero count
Counts while T1
in USR
Counts while T1
in OS or USR
Counts while T1
in OS
T0_OS/T0_USR
= 01
Counts while T0
in USR
Counts while T0
in USR or T1 in
USR
Counts while (a)
T0 in USR or (b)
T1 in OS or (c) T1
in USR
Counts while (a)
T0 in OS or (b) T1
in OS
T0_OS/T0_USR
= 11
Counts while T0
in OS or USR
Counts while (a)
T0 in OS or (b) T0
in USR or (c) T1 in
USR
Counts
irrespective of
CPL, T0, T1
Counts while (a)
T0 in OS or (b) or
T0 in USR or (c)
T1 in OS
T0_OS/T0_USR
= 10
Counts T0 in OS
Counts T0 in OS
or T1 in USR
Counts while
(a)T0 in Os or (b)
T1 in OS or (c) T1
in USR
Counts while (a)
T0 in OS or (b) T1
in OS
When a bit in the event mask field is TI, the effect of specifying bit-0-3 of the associated ESCR are described in Table 15-6. For events that are marked as TI in Appendix
A, the effect of selectively specifying T0_USR, T0_OS, T1_USR, T1_OS bits is shown
in Table 30-32.
30-94 Vol. 3B
PERFORMANCE MONITORING
Table 30-32. Effect of Logical Processor and CPL Qualification
for Non-logical-Processor-specific (TI) Events
T1_OS/T1_USR =
00
T1_OS/T1_USR =
01
T1_OS/T1_USR =
11
T1_OS/T1_USR =
10
T0_OS/T0_USR =
00
Zero count
Counts while (a)
T0 in USR or (b)
T1 in USR
Counts
irrespective of
CPL, T0, T1
Counts while (a)
T0 in OS or (b) T1
in OS
T0_OS/T0_USR =
01
Counts while (a)
T0 in USR or (b)
T1 in USR
Counts while (a)
T0 in USR or (b)
T1 in USR
Counts
irrespective of
CPL, T0, T1
Counts
irrespective of
CPL, T0, T1
T0_OS/T0_USR =
11
Counts
irrespective of
CPL, T0, T1
Counts
irrespective of
CPL, T0, T1
Counts
irrespective of
CPL, T0, T1
Counts
irrespective of
CPL, T0, T1
T0_OS/T0_USR =
0
Counts while (a)
T0 in OS or (b) T1
in OS
Counts
irrespective of
CPL, T0, T1
Counts
irrespective of
CPL, T0, T1
Counts while (a)
T0 in OS or (b) T1
in OS
30.11
COUNTING CLOCKS
The count of cycles, also known as clockticks, forms a the basis for measuring how
long a program takes to execute. Clockticks are also used as part of efficiency ratios
like cycles per instruction (CPI). Processor clocks may stop ticking under circumstances like the following:
•
The processor is halted when there is nothing for the CPU to do. For example, the
processor may halt to save power while the computer is servicing an I/O request.
When Intel Hyper-Threading Technology is enabled, both logical processors must
be halted for performance-monitoring counters to be powered down.
•
The processor is asleep as a result of being halted or because of a powermanagement scheme. There are different levels of sleep. In the some deep sleep
levels, the time-stamp counter stops counting.
In addition, processor core clocks may undergo transitions at different ratios relative
to the processor’s bus clock frequency. Some of the situations that can cause
processor core clock to undergo frequency transitions include:
•
•
TM2 transitions
Enhanced Intel SpeedStep Technology transitions (P-state transitions)
For Intel processors that support Intel Dynamic Acceleration or XE operation, the
processor core clocks may operate at a frequency that differs from the maximum
qualified frequency (as indicated by brand string information reported by CPUID
instruction). See Section 30.11.5 for more detail.
Vol. 3B 30-95
PERFORMANCE MONITORING
There are several ways to count processor clock cycles to monitor performance.
These are:
•
Non-halted clockticks — Measures clock cycles in which the specified logical
processor is not halted and is not in any power-saving state. When Intel HyperThreading Technology is enabled, ticks can be measured on a per-logicalprocessor basis. There are also performance events on dual-core processors that
measure clockticks per logical processor when the processor is not halted.
•
Non-sleep clockticks — Measures clock cycles in which the specified physical
processor is not in a sleep mode or in a power-saving state. These ticks cannot be
measured on a logical-processor basis.
•
Time-stamp counter — Measures clock cycles in which the physical processor is
not in deep sleep. These ticks cannot be measured on a logical-processor basis.
•
Reference clockticks — TM2 or Enhanced Intel SpeedStep technology are two
examples of processor features that can cause processor core clockticks to
represent non-uniform tick intervals due to change of bus ratios. Performance
events that counts clockticks of a constant reference frequency was introduced
Intel Core Duo and Intel Core Solo processors. The mechanism is further
enhanced on processors based on Intel Core microarchitecture.
Some processor models permit clock cycles to be measured when the physical
processor is not in deep sleep (by using the time-stamp counter and the RDTSC
instruction). Note that such ticks cannot be measured on a per-logical-processor
basis. See Section 16.12, “Time-Stamp Counter,” for detail on processor capabilities.
The first two methods use performance counters and can be set up to cause an interrupt upon overflow (for sampling). They may also be useful where it is easier for a
tool to read a performance counter than to use a time stamp counter (the timestamp
counter is accessed using the RDTSC instruction).
For applications with a significant amount of I/O, there are two ratios of interest:
•
Non-halted CPI — Non-halted clockticks/instructions retired measures the CPI
for phases where the CPU was being used. This ratio can be measured on a
logical-processor basis when Intel Hyper-Threading Technology is enabled.
•
Nominal CPI — Time-stamp counter ticks/instructions retired measures the CPI
over the duration of a program, including those periods when the machine halts
while waiting for I/O.
30.11.1 Non-Halted Clockticks
Use the following procedure to program ESCRs and CCCRs to obtain non-halted
clockticks on processors based on Intel NetBurst microarchitecture:
1. Select an ESCR for the global_power_events and specify the RUNNING sub-event
mask and the desired T0_OS/T0_USR/T1_OS/T1_USR bits for the targeted
processor.
30-96 Vol. 3B
PERFORMANCE MONITORING
2. Select an appropriate counter.
3. Enable counting in the CCCR for that counter by setting the enable bit.
30.11.2 Non-Sleep Clockticks
Performance monitoring counters can be configured to count clockticks whenever the
performance monitoring hardware is not powered-down. To count Non-sleep Clockticks with a performance-monitoring counter, do the following:
1. Select one of the 18 counters.
2. Select any of the ESCRs whose events the selected counter can count. Set its
event select to anything other than no_event. This may not seem necessary, but
the counter may be disabled if this is not done.
3. Turn threshold comparison on in the CCCR by setting the compare bit to 1.
4. Set the threshold to 15 and the complement to 1 in the CCCR. Since no event can
exceed this threshold, the threshold condition is met every cycle and the counter
counts every cycle. Note that this overrides any qualification (e.g. by CPL)
specified in the ESCR.
5. Enable counting in the CCCR for the counter by setting the enable bit.
In most cases, the counts produced by the non-halted and non-sleep metrics are
equivalent if the physical package supports one logical processor and is not placed in
a power-saving state. Operating systems may execute an HLT instruction and place a
physical processor in a power-saving state.
On processors that support Intel Hyper-Threading Technology (Intel HT Technology),
each physical package can support two or more logical processors. Current implementation of Intel HT Technology provides two logical processors for each physical
processor. While both logical processors can execute two threads simultaneously,
one logical processor may halt to allow the other logical processor to execute without
sharing execution resources between two logical processors.
Non-halted Clockticks can be set up to count the number of processor clock cycles for
each logical processor whenever the logical processor is not halted (the count may
include some portion of the clock cycles for that logical processor to complete a transition to a halted state). Physical processors that support Intel HT Technology enter
into a power-saving state if all logical processors halt.
The Non-sleep Clockticks mechanism uses a filtering mechanism in CCCRs. The
mechanism will continue to increment as long as one logical processor is not halted
or in a power-saving state. Applications may cause a processor to enter into a powersaving state by using an OS service that transfers control to an OS’s idle loop. The
idle loop then may place the processor into a power-saving state after an implementation-dependent period if there is no work for the processor.
Vol. 3B 30-97
PERFORMANCE MONITORING
30.11.3 Incrementing the Time-Stamp Counter
The time-stamp counter increments when the clock signal on the system bus is
active and when the sleep pin is not asserted. The counter value can be read with the
RDTSC instruction.
The time-stamp counter and the non-sleep clockticks count may not agree in all
cases and for all processors. See Section 16.12, “Time-Stamp Counter,” for more
information on counter operation.
30.11.4 Non-Halted Reference Clockticks
Software can use either processor-specific performance monitor events (for
example: CPU_CLK_UNHALTED.BUS on processors based on the Intel Core microarchitecture, and equivalent event specifications on the Intel Core Duo and Intel Core
Solo processors) to count non-halted reference clockticks.
These events count reference clock cycles whenever the specified processor is not
halted. The counter counts reference cycles associated with a fixed-frequency clock
source irrespective of P-state, TM2, or frequency transitions that may occur to the
processor.
30.11.5 Cycle Counting and Opportunistic Processor Operation
As a result of the state transitions due to opportunistic processor performance operation (see Chapter 14, “Power and Thermal Management”), a logical processor or a
processor core can operate at frequency different from that indicated by the
processor’s maximum qualified frequency.
The following items are expected to hold true irrespective of when opportunistic
processor operation causes state transitions:
•
•
The time stamp counter operates at a fixed-rate frequency of the processor.
•
The IA32_FIXED_CTR2 counter increments at the same TSC frequency
irrespective of any transitions caused by opportunistic processor operation.
•
The Local APIC timer operation is unaffected by opportunistic processor
operation.
•
The TSC, IA32_MPERF, and IA32_FIXED_CTR2 operate at the same, maximumresolved frequency of the platform, which is equal to the product of scalable bus
frequency and maximum resolved bus ratio.
The IA32_MPERF counter increments at the same TSC frequency irrespective of
any transitions caused by opportunistic processor operation.
For processors based on Intel Core microarchitecture, the scalable bus frequency is
encoded in the bit field MSR_FSB_FREQ[2:0] at (0CDH), see Appendix B, “Model-
30-98 Vol. 3B
PERFORMANCE MONITORING
Specific Registers (MSRs)”. The maximum resolved bus ratio can be read from the
following bit field:
•
If XE operation is disabled, the maximum resolved bus ratio can be read in
MSR_PLATFORM_ID[12:8]. It corresponds to the maximum qualified frequency.
•
IF XE operation is enabled, the maximum resolved bus ratio is given in
MSR_PERF_STAT[44:40], it corresponds to the maximum XE operation
frequency configured by BIOS.
XE operation of an Intel 64 processor is implementation specific. XE operation can be
enabled only by BIOS. If MSR_PERF_STAT[31] is set, XE operation is enabled. The
MSR_PERF_STAT[31] field is read-only.
30.12
PERFORMANCE MONITORING, BRANCH PROFILING
AND SYSTEM EVENTS
When performance monitoring facilities and/or branch profiling facilities (see Section
16.5, “Last Branch, Interrupt, and Exception Recording (Intel® Core™2 Duo and
Intel® Atom™ Processor Family)”) are enabled, these facilities capture event counts,
branch records and branch trace messages occurring in a logical processor. The
occurrence of interrupts, instruction streams due to various interrupt handlers all
contribute to the results recorded by these facilities.
If CPUID.01H:ECX.PDCM[bit 15] is 1, the processor supports the
IA32_PERF_CAPABILITIES MSR. If
IA32_PERF_CAPABILITIES.FREEZE_WHILE_SMM[Bit 12] is 1, the processor supports
the ability for system software using performance monitoring and/or branch profiling
facilities to filter out the effects of servicing system management interrupts.
If the FREEZE_WHILE_SMM capability is enabled on a logical processor and after an
SMI is delivered, the processor will clear all the enable bits of
IA32_PERF_GLOBAL_CTRL, save a copy of the content of IA32_DEBUGCTL and
disable LBR, BTF, TR, and BTS fields of IA32_DEBUGCTL before transferring control to
the SMI handler.
The enable bits of IA32_PERF_GLOBAL_CTRL will be set to 1, the saved copy of
IA32_DEBUGCTL prior to SMI delivery will be restored , after the SMI handler issues
RSM to complete its servicing.
It is the responsibility of the SMM code to ensure the state of the performance monitoring and branch profiling facilities are preserved upon entry or until prior to exiting
the SMM. If any of this state is modified due to actions by the SMM code, the SMM
code is required to restore such state to the values present at entry to the SMM
handler.
System software is allowed to set IA32_DEBUGCTL.FREEZE_WHILE_SMM_EN[bit 14]
to 1 only supported as indicated by
IA32_PERF_CAPABILITIES.FREEZE_WHILE_SMM[Bit 12] reporting 1.
Vol. 3B 30-99
PERFORMANCE MONITORING
63
13 12 11
8 7 6 5 43 2 1 0
FW_WRITE (R/O)
SMM_FREEZE (R/O)
PEBS_REC_FMT (R/O)
PEBS_ARCH_REG (R/O)
PEBS_TRAP (R/O)
LBR_FMT (R/O) - 0: 32bit, 1: 64-bit LIP, 2: 64bit EIP
Reserved
Figure 30-39. Layout of IA32_PERF_CAPABILITIES MSR
30.13
PERFORMANCE MONITORING AND DUAL-CORE
TECHNOLOGY
The performance monitoring capability of dual-core processors duplicates the
microarchitectural resources of a single-core processor implementation. Each
processor core has dedicated performance monitoring resources.
In the case of Pentium D processor, each logical processor is associated with dedicated resources for performance monitoring. In the case of Pentium processor
Extreme edition, each processor core has dedicated resources, but two logical
processors in the same core share performance monitoring resources (see Section
30.10, “Performance Monitoring and Intel Hyper-Threading Technology in Processors
Based on Intel NetBurst® Microarchitecture”).
30.14
PERFORMANCE MONITORING ON 64-BIT INTEL XEON
PROCESSOR MP WITH UP TO 8-MBYTE L3 CACHE
The 64-bit Intel Xeon processor MP with up to 8-MByte L3 cache has a CPUID signature of family [0FH], model [03H or 04H]. Performance monitoring capabilities available to Pentium 4 and Intel Xeon processors with the same values (see Section 30.1
and Section 30.10) apply to the 64-bit Intel Xeon processor MP with an L3 cache.
The level 3 cache is connected between the system bus and IOQ through additional
control logic. See Figure 30-40.
30-100 Vol. 3B
PERFORMANCE MONITORING
6\VWHP%XV
L%864DQGL6134
UG/HYHO&DFKH
RUZD\
L)6%
,24
3URFHVVRU&RUH
)URQWHQG([HFXWLRQ
5HWLUHPHQW//
Figure 30-40. Block Diagram of 64-bit Intel Xeon Processor MP with 8-MByte L3
Additional performance monitoring capabilities and facilities unique to 64-bit Intel
Xeon processor MP with an L3 cache are described in this section. The facility for
monitoring events consists of a set of dedicated model-specific registers (MSRs),
each dedicated to a specific event. Programming of these MSRs requires using
RDMSR/WRMSR instructions with 64-bit values.
The lower 32-bits of the MSRs at addresses 107CC through 107D3 are treated as 32
bit performance counter registers. These performance counters can be accessed
using RDPMC instruction with the index starting from 18 through 25. The EDX
register returns zero when reading these 8 PMCs.
The performance monitoring capabilities consist of four events. These are:
•
IBUSQ event — This event detects the occurrence of micro-architectural
conditions related to the iBUSQ unit. It provides two MSRs: MSR_IFSB_IBUSQ0
and MSR_IFSB_IBUSQ1. Configure sub-event qualification and enable/disable
functions using the high 32 bits of these MSRs. The low 32 bits act as a 32-bit
event counter. Counting starts after software writes a non-zero value to one or
more of the upper 32 bits. See Figure 30-41.
Vol. 3B 30-101
PERFORMANCE MONITORING
Reserved
MSR_IFSB_IBUSQx, Addresses: 107CCH and 107CDH
63
60 59 58 57 56 55
49 48
46 45
38 37 36 35 34 33 32
1 1
Saturate
Fill_match
Eviction_match
L3_state_match
Snoop_match
Type_match
T1_match
T0_match
31
0
32 bit event count
Figure 30-41. MSR_IFSB_IBUSQx, Addresses: 107CCH and 107CDH
•
ISNPQ event — This event detects the occurrence of microarchitectural
conditions related to the iSNPQ unit. It provides two MSRs: MSR_IFSB_ISNPQ0
and MSR_IFSB_ISNPQ1. Configure sub-event qualifications and enable/disable
functions using the high 32 bits of the MSRs. The low 32-bits act as a 32-bit event
counter. Counting starts after software writes a non-zero value to one or more of
the upper 32-bits. See Figure 30-42.
30-102 Vol. 3B
PERFORMANCE MONITORING
MSR_IFSB_ISNPQx, Addresses: 107CEH and 107CFH
63
60 59 58 57 56 55
48
46 45
Reserved
39 38 37 36 35 34 33 32
Saturate
L3_state_match
Snoop_match
Type_match
Agent_match
T1_match
T0_match
31
0
32 bit event count
Figure 30-42. MSR_IFSB_ISNPQx, Addresses: 107CEH and 107CFH
•
EFSB event — This event can detect the occurrence of micro-architectural
conditions related to the iFSB unit or system bus. It provides two MSRs:
MSR_EFSB_DRDY0 and MSR_EFSB_DRDY1. Configure sub-event qualifications
and enable/disable functions using the high 32 bits of the 64-bit MSR. The low
32-bit act as a 32-bit event counter. Counting starts after software writes a nonzero value to one or more of the qualification bits in the upper 32-bits of the MSR.
See Figure 30-43.
Vol. 3B 30-103
PERFORMANCE MONITORING
MSR_EFSB_DRDYx, Addresses: 107D0H and 107D1H
63
60 59 58 57 56 55
50 49 48
Reserved
39 38 37 36 35 34 33 32
Saturate
Other
Own
31
0
32 bit event count
Figure 30-43. MSR_EFSB_DRDYx, Addresses: 107D0H and 107D1H
•
IBUSQ Latency event — This event accumulates weighted cycle counts for
latency measurement of transactions in the iBUSQ unit. The count is enabled by
setting MSR_IFSB_CTRL6[bit 26] to 1; the count freezes after software sets
MSR_IFSB_CTRL6[bit 26] to 0. MSR_IFSB_CNTR7 acts as a 64-bit event
counter for this event. See Figure 30-44.
30-104 Vol. 3B
PERFORMANCE MONITORING
MSR_IFSB_CTL6 Address: 107D2H
63
59
0
57
Enable
Reserved
MSR_IFSB_CNTR7 Address: 107D3H
0
63
64 bit event count
Figure 30-44. MSR_IFSB_CTL6, Address: 107D2H;
MSR_IFSB_CNTR7, Address: 107D3H
30.15
PERFORMANCE MONITORING ON L3 AND CACHING
BUS CONTROLLER SUB-SYSTEMS
The Intel Xeon processor 7400 series and Dual-Core Intel Xeon processor 7100
series employ a distinct L3/caching bus controller sub-system. These sub-system
have a unique set of performance monitoring capability and programming interfaces
that are largely common between these two processor families.
Intel Xeon processor 7400 series are based on 45nm enhanced Intel Core microarchitecture. The CPUID signature is indicated by DisplayFamily_DisplayModel value of
06_1DH (see CPUID instruction in Chapter 3, “Instruction Set Reference, A-M” in the
Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 2A). Intel
Xeon processor 7400 series have six processor cores that share an L3 cache.
Dual-Core Intel Xeon processor 7100 series are based on Intel NetBurst microarchitecture, have a CPUID signature of family [0FH], model [06H] and a unified L3 cache
shared between two cores. Each core in an Intel Xeon processor 7100 series supports
Intel Hyper-Threading Technology, providing two logical processors per core.
Both Intel Xeon processor 7400 series and Intel Xeon processor 7100 series support
multi-processor configurations using system bus interfaces. In Intel Xeon processor
7400 series, the L3/caching bus controller sub-system provides three Simple Direct
Interface (SDI) to service transactions originated the XQ-replacement SDI logic in
each dual-core modules. In Intel Xeon processor 7100 series, the IOQ logic in each
processor core is replaced with a Simple Direct Interface (SDI) logic. The L3 cache is
Vol. 3B 30-105
PERFORMANCE MONITORING
connected between the system bus and the SDI through additional control logic. See
Figure 30-45 for the block configuration of six processor cores and the L3/Caching
bus controller sub-system in Intel Xeon processor 7400 series. Figure 30-45 shows
the block configuration of two processor cores (four logical processors) and the
L3/Caching bus controller sub-system in Intel Xeon processor 7100 series.
FSB
GBSQ, GSNPQ,
GINTQ, ...
L3
SDI
SDI interface
L2
Core
SDI interface
L2
Core
Core
SDI interface
L2
Core
Core
Core
Figure 30-45. Block Diagram of Intel Xeon Processor 7400 Series
Almost all of the performance monitoring capabilities available to processor cores
with the same CPUID signatures (see Section 30.1 and Section 30.10) apply to Intel
Xeon processor 7100 series. The MSRs used by performance monitoring interface are
shared between two logical processors in the same processor core.
The performance monitoring capabilities available to processor with
DisplayFamily_DisplayModel signature 06_17H also apply to Intel Xeon processor
7400 series. Each processor core provides its own set of MSRs for performance monitoring interface.
The IOQ_allocation and IOQ_active_entries events are not supported in Intel Xeon
processor 7100 series and 7400 series. Additional performance monitoring capabilities applicable to the L3/caching bus controller sub-system are described in this
section.
30-106 Vol. 3B
PERFORMANCE MONITORING
FSB
GBSQ, GSNPQ,
GINTQ, ...
L3
SDI
SDI interface
SDI interface
Processor core
Processor core
Logical
processor
Logical
processor
Logical
processor
Logical
processor
Figure 30-46. Block Diagram of Intel Xeon Processor 7100 Series
30.15.1 Overview of Performance Monitoring with L3/Caching Bus
Controller
The facility for monitoring events consists of a set of dedicated model-specific
registers (MSRs). There are eight event select/counting MSRs that are dedicated to
counting events associated with specified microarchitectural conditions. Programming of these MSRs requires using RDMSR/WRMSR instructions with 64-bit values.
In addition, an MSR MSR_EMON_L3_GL_CTL provides simplified interface to control
freezing, resetting, re-enabling operation of any combination of these event
select/counting MSRs.
The eight MSRs dedicated to count occurrences of specific conditions are further
divided to count three sub-classes of microarchitectural conditions:
•
Two MSRs (MSR_EMON_L3_CTR_CTL0 and MSR_EMON_L3_CTR_CTL1) are
dedicated to counting GBSQ events. Up to two GBSQ events can be programmed
and counted simultaneously.
•
Two MSRs (MSR_EMON_L3_CTR_CTL2 and MSR_EMON_L3_CTR_CTL3) are
dedicated to counting GSNPQ events. Up to two GBSQ events can be
programmed and counted simultaneously.
Vol. 3B 30-107
PERFORMANCE MONITORING
•
Four MSRs (MSR_EMON_L3_CTR_CTL4, MSR_EMON_L3_CTR_CTL5,
MSR_EMON_L3_CTR_CTL6, and MSR_EMON_L3_CTR_CTL7) are dedicated to
counting external bus operations.
The bit fields in each of eight MSRs share the following common characteristics:
•
Bits 63:32 is the event control field that includes an event mask and other bit
fields that control counter operation. The event mask field specifies details of the
microarchitectural condition, and its definition differs across GBSQ, GSNPQ, FSB.
•
Bits 31:0 is the event count field. If the specified condition is met during each
relevant clock domain of the event logic, the matched condition signals the
counter logic to increment the associated event count field. The lower 32-bits of
these 8 MSRs at addresses 107CC through 107D3 are treated as 32 bit
performance counter registers.
In Dual-Core Intel Xeon processor 7100 series, the uncore performance counters can
be accessed using RDPMC instruction with the index starting from 18 through 25. The
EDX register returns zero when reading these 8 PMCs.
In Intel Xeon processor 7400 series, RDPMC with ECX between 2 and 9 can be used
to access the eight uncore performance counter/control registers.
30.15.2 GBSQ Event Interface
The layout of MSR_EMON_L3_CTR_CTL0 and MSR_EMON_L3_CTR_CTL1 is given in
Figure 30-47. Counting starts after software writes a non-zero value to one or more
of the upper 32 bits.
The event mask field (bits 58:32) consists of the following eight attributes:
•
Agent_Select (bits 35:32): The definition of this field differs slightly between
Intel Xeon processor 7100 and 7400.
For Intel Xeon processor 7100 series, each bit specifies a logical processor in the
physical package. The lower two bits corresponds to two logical processors in the
first processor core, the upper two bits corresponds to two logical processors in
the second processor core. 0FH encoding matches transactions from any logical
processor.
For Intel Xeon processor 7400 series, each bit of [34:32] specifies the SDI logic
of a dual-core module as the originator of the transaction. A value of 0111B in
bits [35:32] specifies transaction from any processor core.
30-108 Vol. 3B
PERFORMANCE MONITORING
MSR_EMON_L3_CTR_CTL0/1, Addresses: 107CCH/107CDH
63
60 59 58 57 56 55 54 53
47 46
44 43
Reserved
38 37 36 35
32
Saturate
Cross_snoop
Fill_eviction
Core_module_select
L3_state
Snoop_match
Type_match
Data_flow
Agent_select
31
0
32 bit event count
Figure 30-47. MSR_EMON_L3_CTR_CTL0/1, Addresses: 107CCH/107CDH
•
Data_Flow (bits 37:36): Bit 36 specifies demand transactions, bit 37 specifies
prefetch transactions.
•
Type_Match (bits 43:38): Specifies transaction types. If all six bits are set, event
count will include all transaction types.
•
Snoop_Match: (bits 46:44): The three bits specify (in ascending bit position)
clean snoop result, HIT snoop result, and HITM snoop results respectively.
•
•
L3_State (bits 53:47): Each bit specifies an L2 coherency state.
Core_Module_Select (bits 55:54): The valid encodings for L3 lookup differ
slightly between Intel Xeon processor 7100 and 7400.
For Intel Xeon processor 7100 series,
— 00B: Match transactions from any core in the physical package
— 01B: Match transactions from this core only
— 10B: Match transactions from the other core in the physical package
— 11B: Match transaction from both cores in the physical package
For Intel Xeon processor 7400 series,
— 00B: Match transactions from any dual-core module in the physical package
Vol. 3B 30-109
PERFORMANCE MONITORING
— 01B: Match transactions from this dual-core module only
— 10B: Match transactions from either one of the other two dual-core modules
in the physical package
— 11B: Match transaction from more than one dual-core modules in the
physical package
•
Fill_Eviction (bits 57:56): The valid encodings are
— 00B: Match any transactions
— 01B: Match transactions that fill L3
— 10B: Match transactions that fill L3 without an eviction
— 11B: Match transaction fill L3 with an eviction
•
Cross_Snoop (bit 58): The encodings are
\
— 0B: Match any transactions
— 1B: Match cross snoop transactions
For each counting clock domain, if all eight attributes match, event logic signals to
increment the event count field.
30.15.3 GSNPQ Event Interface
The layout of MSR_EMON_L3_CTR_CTL2 and MSR_EMON_L3_CTR_CTL3 is given in
Figure 30-48. Counting starts after software writes a non-zero value to one or more
of the upper 32 bits.
The event mask field (bits 58:32) consists of the following six attributes:
•
Agent_Select (bits 37:32): The definition of this field differs slightly between
Intel Xeon processor 7100 and 7400.
•
For Intel Xeon processor 7100 series, each of the lowest 4 bits specifies a logical
processor in the physical package. The lowest two bits corresponds to two logical
processors in the first processor core, the next two bits corresponds to two logical
processors in the second processor core. Bit 36 specifies other symmetric agent
transactions. Bit 37 specifies central agent transactions. 3FH encoding matches
transactions from any logical processor.
For Intel Xeon processor 7400 series, each of the lowest 3 bits specifies a dualcore module in the physical package. Bit 37 specifies central agent transactions.
•
Type_Match (bits 43:38): Specifies transaction types. If all six bits are set, event
count will include any transaction types.
•
Snoop_Match: (bits 46:44): The three bits specify (in ascending bit position)
clean snoop result, HIT snoop result, and HITM snoop results respectively.
•
•
L2_State (bits 53:47): Each bit specifies an L3 coherency state.
Core_Module_Select (bits 56:54): Bit 56 enables Core_Module_Select matching.
If bit 56 is clear, Core_Module_Select encoding is ignored. The valid encodings for
30-110 Vol. 3B
PERFORMANCE MONITORING
the lower two bits (bit 55, 54) differ slightly between Intel Xeon processor 7100
and 7400.
For Intel Xeon processor 7100 series, if bit 56 is set, the valid encodings for the
lower two bits (bit 55, 54) are
— 00B: Match transactions from only one core (irrespective which core) in the
physical package
— 01B: Match transactions from this core and not the other core
— 10B: Match transactions from the other core in the physical package, but not
this core
— 11B: Match transaction from both cores in the physical package
For Intel Xeon processor 7400 series, if bit 56 is set, the valid encodings for the
lower two bits (bit 55, 54) are
— 00B: Match transactions from only one dual-core module (irrespective which
module) in the physical package
— 01B: Match transactions from one or more dual-core modules.
— 10B: Match transactions from two or more dual-core modules.
— 11B: Match transaction from all three dual-core modules in the physical
package
•
Block_Snoop (bit 57): specifies blocked snoop.
For each counting clock domain, if all six attributes match, event logic signals to
increment the event count field.
Vol. 3B 30-111
PERFORMANCE MONITORING
MSR_EMON_L3_CTR_CTL2/3, Addresses: 107CEH/107CFH
63
60 59 58 57 56 55 54 53
47 46
44 43
Reserved
39 38 37 36
32
Saturate
Block_snoop
Core_select
L2_state
Snoop_match
Type_match
Agent_match
31
0
32 bit event count
Figure 30-48. MSR_EMON_L3_CTR_CTL2/3, Addresses: 107CEH/107CFH
30.15.4 FSB Event Interface
The layout of MSR_EMON_L3_CTR_CTL4 through MSR_EMON_L3_CTR_CTL7 is given
in Figure 30-49. Counting starts after software writes a non-zero value to one or
more of the upper 32 bits.
The event mask field (bits 58:32) is organized as follows:
•
•
Bit 58: must set to 1.
FSB_Submask (bits 57:32): Specifies FSB-specific sub-event mask.
The FSB sub-event mask defines a set of independent attributes. The event logic
signals to increment the associated event count field if one of the attribute matches.
Some of the sub-event mask bit counts durations. A duration event increments at
most once per cycle.
30-112 Vol. 3B
PERFORMANCE MONITORING
MSR_EMON_L3_CTR_CTL4/5/6/7, Addresses: 107D0H-107D3H
63
60 59 58 57 56 55
50 49 48
Reserved
39 38 37 36 35 34 33 32
1
Saturate
FSB submask
31
0
32 bit event count
Figure 30-49. MSR_EMON_L3_CTR_CTL4/5/6/7, Addresses: 107D0H-107D3H
30.15.4.1 FSB Sub-Event Mask Interface
•
FSB_type (bit 37:32): Specifies different FSB transaction types originated from
this physical package
•
FSB_L_clear (bit 38): Count clean snoop results from any source for transaction
originated from this physical package
•
FSB_L_hit (bit 39): Count HIT snoop results from any source for transaction
originated from this physical package
•
FSB_L_hitm (bit 40): Count HITM snoop results from any source for transaction
originated from this physical package
•
•
•
•
FSB_L_defer (bit 41): Count DEFER responses to this processor’s transactions
•
•
•
•
•
FSB_DRDY (bit 45): Count DRDY assertions by this processor
FSB_L_retry (bit 42): Count RETRY responses to this processor’s transactions
FSB_L_snoop_stall (bit 43): Count snoop stalls to this processor’s transactions
FSB_DBSY (bit 44): Count DBSY assertions by this processor (without a
concurrent DRDY)
FSB_BNR (bit 46): Count BNR assertions by this processor
FSB_IOQ_empty (bit 47): Counts each bus clocks when the IOQ is empty
FSB_IOQ_full (bit 48): Counts each bus clocks when the IOQ is full
FSB_IOQ_active (bit 49): Counts each bus clocks when there is at least one entry
in the IOQ
Vol. 3B 30-113
PERFORMANCE MONITORING
•
•
FSB_WW_data (bit 50): Counts back-to-back write transaction’s data phase.
•
FSB_WR_issue (bit 52): Counts back-to-back write-read transaction request
pairs issued by this processor.
•
FSB_RW_issue (bit 53): Counts back-to-back read-write transaction request
pairs issued by this processor.
•
FSB_other_DBSY (bit 54): Count DBSY assertions by another agent (without a
concurrent DRDY)
•
•
FSB_other_DRDY (bit 55): Count DRDY assertions by another agent
•
FSB_other_BNR (bit 57): Count BNR assertions from another agent
FSB_WW_issue (bit 51): Counts back-to-back write transaction request pairs
issued by this processor.
FSB_other_snoop_stall (bit 56): Count snoop stalls on the FSB due to another
agent
30.15.5 Common Event Control Interface
The MSR_EMON_L3_GL_CTL MSR provides simplified access to query overflow status
of the GBSQ, GSNPQ, FSB event counters. It also provides control bit fields to freeze,
unfreeze, or reset those counters. The following bit fields are supported:
•
GL_freeze_cmd (bit 0): Freeze the event counters specified by the
GL_event_select field.
•
GL_unfreeze_cmd (bit 1): Unfreeze the event counters specified by the
GL_event_select field.
•
GL_reset_cmd (bit 2): Clear the event count field of the event counters specified
by the GL_event_select field. The event select field is not affected.
•
GL_event_select (bit 23:16): Selects one or more event counters to subject to
specified command operations indicated by bits 2:0. Bit 16 corresponds to
MSR_EMON_L3_CTR_CTL0, bit 23 corresponds to MSR_EMON_L3_CTR_CTL7.
•
GL_event_status (bit 55:48): Indicates the overflow status of each event
counters. Bit 48 corresponds to MSR_EMON_L3_CTR_CTL0, bit 55 corresponds
to MSR_EMON_L3_CTR_CTL7.
In the event control field (bits 63:32) of each MSR, if the saturate control (bit 59, see
Figure 30-47 for example) is set, the event logic forces the value FFFF_FFFFH into
the event count field instead of incrementing it.
30.16
PERFORMANCE MONITORING (P6 FAMILY
PROCESSOR)
The P6 family processors provide two 40-bit performance counters, allowing two
types of events to be monitored simultaneously. These can either count events or
30-114 Vol. 3B
PERFORMANCE MONITORING
measure duration. When counting events, a counter increments each time a specified event takes place or a specified number of events takes place. When measuring
duration, it counts the number of processor clocks that occur while a specified condition is true. The counters can count events or measure durations that occur at any
privilege level.
Table A-22, Appendix A, lists the events that can be counted with the P6 family
performance monitoring counters.
NOTE
The performance-monitoring events listed in Appendix A are
intended to be used as guides for performance tuning. Counter
values reported are not guaranteed to be accurate and should be
used as a relative guide for tuning. Known discrepancies are
documented where applicable.
The performance-monitoring counters are supported by four MSRs: the performance
event select MSRs (PerfEvtSel0 and PerfEvtSel1) and the performance counter MSRs
(PerfCtr0 and PerfCtr1). These registers can be read from and written to using the
RDMSR and WRMSR instructions, respectively. They can be accessed using these
instructions only when operating at privilege level 0. The PerfCtr0 and PerfCtr1 MSRs
can be read from any privilege level using the RDPMC (read performance-monitoring
counters) instruction.
NOTE
The PerfEvtSel0, PerfEvtSel1, PerfCtr0, and PerfCtr1 MSRs and the
events listed in Table A-22 are model-specific for P6 family
processors. They are not guaranteed to be available in other IA-32
processors.
30.16.1 PerfEvtSel0 and PerfEvtSel1 MSRs
The PerfEvtSel0 and PerfEvtSel1 MSRs control the operation of the performancemonitoring counters, with one register used to set up each counter. They specify the
events to be counted, how they should be counted, and the privilege levels at which
counting should take place. Figure 30-50 shows the flags and fields in these MSRs.
The functions of the flags and fields in the PerfEvtSel0 and PerfEvtSel1 MSRs are as
follows:
•
Event select field (bits 0 through 7) — Selects the event logic unit to detect
certain microarchitectural conditions (see Table A-22, for a list of events and their
8-bit codes).
•
Unit mask (UMASK) field (bits 8 through 15) — Further qualifies the event
logic unit selected in the event select field to detect a specific microarchitectural
condition. For example, for some cache events, the mask is used as a MESIprotocol qualifier of cache states (see Table A-22).
Vol. 3B 30-115
PERFORMANCE MONITORING
31
24 23 22 21 20 19 18 17 16 15
Counter Mask
(CMASK)
I
N E
V N
0
8 7
I
U
N P E O S Unit Mask (UMASK)
S R
T C
Event Select
INV—Invert counter mask
EN—Enable counters*
INT—APIC interrupt enable
PC—Pin control
E—Edge detect
OS—Operating system mode
USR—User Mode
* Only available in PerfEvtSel0.
Reserved
Figure 30-50. PerfEvtSel0 and PerfEvtSel1 MSRs
•
USR (user mode) flag (bit 16) — Specifies that events are counted only when
the processor is operating at privilege levels 1, 2 or 3. This flag can be used in
conjunction with the OS flag.
•
OS (operating system mode) flag (bit 17) — Specifies that events are
counted only when the processor is operating at privilege level 0. This flag can be
used in conjunction with the USR flag.
•
E (edge detect) flag (bit 18) — Enables (when set) edge detection of events.
The processor counts the number of deasserted to asserted transitions of any
condition that can be expressed by the other fields. The mechanism is limited in
that it does not permit back-to-back assertions to be distinguished. This
mechanism allows software to measure not only the fraction of time spent in a
particular state, but also the average length of time spent in such a state (for
example, the time spent waiting for an interrupt to be serviced).
•
PC (pin control) flag (bit 19) — When set, the processor toggles the PMi pins
and increments the counter when performance-monitoring events occur; when
clear, the processor toggles the PMi pins when the counter overflows. The
toggling of a pin is defined as assertion of the pin for a single bus clock followed
by deassertion.
•
INT (APIC interrupt enable) flag (bit 20) — When set, the processor
generates an exception through its local APIC on counter overflow.
•
EN (Enable Counters) Flag (bit 22) — This flag is only present in the
PerfEvtSel0 MSR. When set, performance counting is enabled in both
performance-monitoring counters; when clear, both counters are disabled.
•
INV (invert) flag (bit 23) — Inverts the result of the counter-mask comparison
when set, so that both greater than and less than comparisons can be made.
30-116 Vol. 3B
PERFORMANCE MONITORING
•
Counter mask (CMASK) field (bits 24 through 31) — When nonzero, the
processor compares this mask to the number of events counted during a single
cycle. If the event count is greater than or equal to this mask, the counter is
incremented by one. Otherwise the counter is not incremented. This mask can be
used to count events only if multiple occurrences happen per clock (for example,
two or more instructions retired per clock). If the counter-mask field is 0, then
the counter is incremented each cycle by the number of events that occurred that
cycle.
30.16.2 PerfCtr0 and PerfCtr1 MSRs
The performance-counter MSRs (PerfCtr0 and PerfCtr1) contain the event or duration
counts for the selected events being counted. The RDPMC instruction can be used by
programs or procedures running at any privilege level and in virtual-8086 mode to
read these counters. The PCE flag in control register CR4 (bit 8) allows the use of this
instruction to be restricted to only programs and procedures running at privilege
level 0.
The RDPMC instruction is not serializing or ordered with other instructions. Thus, it
does not necessarily wait until all previous instructions have been executed before
reading the counter. Similarly, subsequent instructions may begin execution before
the RDPMC instruction operation is performed.
Only the operating system, executing at privilege level 0, can directly manipulate the
performance counters, using the RDMSR and WRMSR instructions. A secure operating system would clear the PCE flag during system initialization to disable direct
user access to the performance-monitoring counters, but provide a user-accessible
programming interface that emulates the RDPMC instruction.
The WRMSR instruction cannot arbitrarily write to the performance-monitoring
counter MSRs (PerfCtr0 and PerfCtr1). Instead, the lower-order 32 bits of each MSR
may be written with any value, and the high-order 8 bits are sign-extended according
to the value of bit 31. This operation allows writing both positive and negative values
to the performance counters.
30.16.3 Starting and Stopping the Performance-Monitoring Counters
The performance-monitoring counters are started by writing valid setup information
in the PerfEvtSel0 and/or PerfEvtSel1 MSRs and setting the enable counters flag in
the PerfEvtSel0 MSR. If the setup is valid, the counters begin counting following the
execution of a WRMSR instruction that sets the enable counter flag. The counters can
be stopped by clearing the enable counters flag or by clearing all the bits in the
PerfEvtSel0 and PerfEvtSel1 MSRs. Counter 1 alone can be stopped by clearing the
PerfEvtSel1 MSR.
Vol. 3B 30-117
PERFORMANCE MONITORING
30.16.4 Event and Time-Stamp Monitoring Software
To use the performance-monitoring counters and time-stamp counter, the operating
system needs to provide an event-monitoring device driver. This driver should
include procedures for handling the following operations:
•
•
•
•
•
Feature checking
Initialize and start counters
Stop counters
Read the event counters
Read the time-stamp counter
The event monitor feature determination procedure must check whether the current
processor supports the performance-monitoring counters and time-stamp counter.
This procedure compares the family and model of the processor returned by the
CPUID instruction with those of processors known to support performance monitoring. (The Pentium and P6 family processors support performance counters.) The
procedure also checks the MSR and TSC flags returned to register EDX by the CPUID
instruction to determine if the MSRs and the RDTSC instruction are supported.
The initialize and start counters procedure sets the PerfEvtSel0 and/or PerfEvtSel1
MSRs for the events to be counted and the method used to count them and initializes
the counter MSRs (PerfCtr0 and PerfCtr1) to starting counts. The stop counters
procedure stops the performance counters (see Section 30.16.3, “Starting and Stopping the Performance-Monitoring Counters”).
The read counters procedure reads the values in the PerfCtr0 and PerfCtr1 MSRs, and
a read time-stamp counter procedure reads the time-stamp counter. These procedures would be provided in lieu of enabling the RDTSC and RDPMC instructions that
allow application code to read the counters.
30.16.5 Monitoring Counter Overflow
The P6 family processors provide the option of generating a local APIC interrupt when
a performance-monitoring counter overflows. This mechanism is enabled by setting
the interrupt enable flag in either the PerfEvtSel0 or the PerfEvtSel1 MSR. The
primary use of this option is for statistical performance sampling.
To use this option, the operating system should do the following things on the
processor for which performance events are required to be monitored:
•
•
Provide an interrupt vector for handling the counter-overflow interrupt.
•
Provide an entry in the IDT that points to a stub exception handler that returns
without executing any instructions.
•
Provide an event monitor driver that provides the actual interrupt handler and
modifies the reserved IDT entry to point to its interrupt routine.
Initialize the APIC PERF local vector entry to enable handling of performancemonitor counter overflow events.
30-118 Vol. 3B
PERFORMANCE MONITORING
When interrupted by a counter overflow, the interrupt handler needs to perform the
following actions:
•
Save the instruction pointer (EIP register), code-segment selector, TSS segment
selector, counter values and other relevant information at the time of the
interrupt.
•
Reset the counter to its initial setting and return from the interrupt.
An event monitor application utility or another application program can read the
information collected for analysis of the performance of the profiled application.
30.17
PERFORMANCE MONITORING (PENTIUM
PROCESSORS)
The Pentium processor provides two 40-bit performance counters, which can be used
to count events or measure duration. The counters are supported by three MSRs: the
control and event select MSR (CESR) and the performance counter MSRs (CTR0 and
CTR1). These can be read from and written to using the RDMSR and WRMSR instructions, respectively. They can be accessed using these instructions only when operating at privilege level 0.
Each counter has an associated external pin (PM0/BP0 and PM1/BP1), which can be
used to indicate the state of the counter to external hardware.
NOTES
The CESR, CTR0, and CTR1 MSRs and the events listed in Table A-23
are model-specific for the Pentium processor.
The performance-monitoring events listed in Appendix A are
intended to be used as guides for performance tuning. Counter
values reported are not guaranteed to be accurate and should be
used as a relative guide for tuning. Known discrepancies are
documented where applicable.
30.17.1
Control and Event Select Register (CESR)
The 32-bit control and event select MSR (CESR) controls the operation of performance-monitoring counters CTR0 and CTR1 and the associated pins (see
Figure 30-51). To control each counter, the CESR register contains a 6-bit event
select field (ES0 and ES1), a pin control flag (PC0 and PC1), and a 3-bit counter
control field (CC0 and CC1). The functions of these fields are as follows:
•
ES0 and ES1 (event select) fields (bits 0-5, bits 16-21) — Selects (by
entering an event code in the field) up to two events to be monitored. See Table
A-23 for a list of available event codes.
Vol. 3B 30-119
PERFORMANCE MONITORING
31
26 25 24
P
C
1
CC1
22 21
16 15
10 9 8
ES1
P
C
0
6 5
CC0
0
ESO
PC1—Pin control 1
CC1—Counter control 1
ES1—Event select 1
PC0—Pin control 0
CC0—Counter control 0
ES0—Event select 0
Reserved
Figure 30-51. CESR MSR (Pentium Processor Only)
•
CC0 and CC1 (counter control) fields (bits 6-8, bits 22-24) — Controls the
operation of the counter. Control codes are as follows:
000 — Count nothing (counter disabled)
001 — Count the selected event while CPL is 0, 1, or 2
010 — Count the selected event while CPL is 3
011 — Count the selected event regardless of CPL
100 — Count nothing (counter disabled)
101 — Count clocks (duration) while CPL is 0, 1, or 2
110 — Count clocks (duration) while CPL is 3
111 — Count clocks (duration) regardless of CPL
The highest order bit selects between counting events and counting clocks
(duration); the middle bit enables counting when the CPL is 3; and the low-order
bit enables counting when the CPL is 0, 1, or 2.
•
PC0 and PC1 (pin control) flags (bits 9, 25) — Selects the function of the
external performance-monitoring counter pin (PM0/BP0 and PM1/BP1). Setting
one of these flags to 1 causes the processor to assert its associated pin when the
counter has overflowed; setting the flag to 0 causes the pin to be asserted when
the counter has been incremented. These flags permit the pins to be individually
programmed to indicate the overflow or incremented condition. The external
signalling of the event on the pins will lag the internal event by a few clocks as the
signals are latched and buffered.
While a counter need not be stopped to sample its contents, it must be stopped and
cleared or preset before switching to a new event. It is not possible to set one
counter separately. If only one event needs to be changed, the CESR register must
30-120 Vol. 3B
PERFORMANCE MONITORING
be read, the appropriate bits modified, and all bits must then be written back to
CESR. At reset, all bits in the CESR register are cleared.
30.17.2
Use of the Performance-Monitoring Pins
When performance-monitor pins PM0/BP0 and/or PM1/BP1 are configured to indicate
when the performance-monitor counter has incremented and an “occurrence event”
is being counted, the associated pin is asserted (high) each time the event occurs.
When a “duration event” is being counted, the associated PM pin is asserted for the
entire duration of the event. When the performance-monitor pins are configured to
indicate when the counter has overflowed, the associated PM pin is asserted when
the counter has overflowed.
When the PM0/BP0 and/or PM1/BP1 pins are configured to signal that a counter has
incremented, it should be noted that although the counters may increment by 1 or 2
in a single clock, the pins can only indicate that the event occurred. Moreover, since
the internal clock frequency may be higher than the external clock frequency, a
single external clock may correspond to multiple internal clocks.
A “count up to” function may be provided when the event pin is programmed to
signal an overflow of the counter. Because the counters are 40 bits, a carry out of bit
39 indicates an overflow. A counter may be preset to a specific value less then 240 −
1. After the counter has been enabled and the prescribed number of events has transpired, the counter will overflow.
Approximately 5 clocks later, the overflow is indicated externally and appropriate
action, such as signaling an interrupt, may then be taken.
The PM0/BP0 and PM1/BP1 pins also serve to indicate breakpoint matches during incircuit emulation, during which time the counter increment or overflow function of
these pins is not available. After RESET, the PM0/BP0 and PM1/BP1 pins are configured for performance monitoring, however a hardware debugger may reconfigure
these pins to indicate breakpoint matches.
30.17.3
Events Counted
Events that performance-monitoring counters can be set to count and record (using
CTR0 and CTR1) are divided in two categories: occurrence and duration:
•
Occurrence events — Counts are incremented each time an event takes place.
If PM0/BP0 or PM1/BP1 pins are used to indicate when a counter increments, the
pins are asserted each clock counters increment. But if an event happens twice in
one clock, the counter increments by 2 (the pins are asserted only once).
•
Duration events — Counters increment the total number of clocks that the
condition is true. When used to indicate when counters increment, PM0/BP0
and/or PM1/BP1 pins are asserted for the duration.
Vol. 3B 30-121
PERFORMANCE MONITORING
30-122 Vol. 3B
APPENDIX A
PERFORMANCE-MONITORING EVENTS
This appendix lists the performance-monitoring events that can be monitored with
the Intel 64 or IA-32 processors. The ability to monitor performance events and the
events that can be monitored in these processors are mostly model-specific, except
for architectural performance events, described in Section A.1.
Non-architectural performance events (i.e. model-specific events) are listed for each
generation of microarchitecture:
•
Section A.2 - Processors based on Intel® microarchitecture code name Sandy
Bridge
•
•
•
•
•
•
•
•
•
•
Section A.3 - Processors based on Intel® microarchitecture code name Nehalem
Section A.4 - Processors based on Intel® microarchitecture code name Westmere
Section A.5 - Processors based on Enhanced Intel® Core™ microarchitecture
Section A.6 - Processors based on Intel® Core™ microarchitecture
Section A.7 - Processors based on Intel® Atom™ microarchitecture
Section A.8 - Intel® Core™ Solo and Intel® Core™ Duo processors
Section A.9 - Processors based on Intel NetBurst® microarchitecture
Section A.10 - Pentium® M family processors
Section A.11 - P6 family processors
Section A.12 - Pentium® processors
NOTE
These performance-monitoring events are intended to be used as
guides for performance tuning. The counter values reported by the
performance-monitoring events are approximate and believed to be
useful as relative guides for tuning software. Known discrepancies
are documented where applicable.
A.1
ARCHITECTURAL PERFORMANCE-MONITORING
EVENTS
Architectural performance events are introduced in Intel Core Solo and Intel Core
Duo processors. They are also supported on processors based on Intel Core microarchitecture. Table A-1 lists pre-defined architectural performance events that can be
configured using general-purpose performance counters and associated event-select
registers.
Vol. 3B A-1
PERFORMANCE-MONITORING EVENTS
Table A-1. Architectural Performance Events
Event
Num.
Event Mask Mnemonic
Umask
Value
3CH
UnHalted Core Cycles
00H
Unhalted core cycles
3CH
UnHalted Reference
Cycles
01H
Unhalted reference cycles
Description
C0H
Instruction Retired
00H
Instruction retired
2EH
LLC Reference
4FH
Last level cache references
2EH
LLC Misses
41H
Last level cache misses
C4H
Branch Instruction Retired 00H
Branch instruction at retirement
C5H
Branch Misses Retired
Mispredicted Branch Instruction at
retirement
00H
Comment
Measures
bus cycle1
NOTES:
1. Implementation of this event in Intel Core 2 processor family, Intel Core Duo, and Intel Core Solo processors measures bus clocks.
A.2
PERFORMANCE MONITORING EVENTS FOR
INTEL® CORE™ PROCESSOR 2XXX SERIES
Second generation Intel® Core™ Processor 2xxx Series are based on the Intel
microarchitecture code name Sandy Bridge. They support the architectural and nonarchitectural performance-monitoring events listed in Table A-1 and Table A-2. The
events in Table A-2 apply to processors with CPUID signature of
DisplayFamily_DisplayModel encoding with the following values: 06_2AH.
Table A-2. Non-Architectural Performance Events In the Processor Core for Intel Core
i7, i5, i3 Processors 2xxx Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
Description
03H
01H
LD_BLOCKS.DATA_U
NKNOWN
blocked loads due to store buffer
blocks with unknown data.
03H
02H
LD_BLOCKS.STORE_F loads blocked by overlapping with
ORWARD
store buffer that cannot be
forwarded .
03H
08H
LD_BLOCKS.NO_SR
03H
10H
LD_BLOCKS.ALL_BLO Number of cases where any load is
CK
blocked but has no DCU miss.
A-2 Vol. 3B
# of Split loads blocked due to
resource not available.
Comment
PERFORMANCE-MONITORING EVENTS
Table A-2. Non-Architectural Performance Events In the Processor Core for Intel Core
i7, i5, i3 Processors 2xxx Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
05H
01H
MISALIGN_MEM_REF. Speculative cache-line split load
LOADS
uops dispatched to L1D.
05H
02H
MISALIGN_MEM_REF. Speculative cache-line split StoreSTORES
address uops dispatched to L1D.
07H
01H
LD_BLOCKS_PARTIA
L.ADDRESS_ALIAS
False dependencies in MOB due to
partial compare on address.
07H
08H
LD_BLOCKS_PARTIA
L.ALL_STA_BLOCK
The number of times that load
operations are temporarily blocked
because of older stores, with
addresses that are not yet known. A
load operation may incur more than
one block of this type.
08H
01H
DTLB_LOAD_MISSES. Misses in all TLB levels that cause a
MISS_CAUSES_A_WA page walk of any page size.
LK
08H
02H
DTLB_LOAD_MISSES. Misses in all TLB levels that caused
WALK_COMPLETED
page walk completed of any size.
08H
04H
DTLB_LOAD_MISSES. Cycle PMH is busy with a walk.
WALK_DURATION
08H
10H
DTLB_LOAD_MISSES. Number of cache load STLB hits. No
STLB_HIT
page walk.
0DH
03H
INT_MISC.RECOVERY
_CYCLES
0DH
40H
INT_MISC.RAT_STALL Cycles RAT external stall is sent to
_CYCLES
IDQ for this thread.
0EH
01H
UOPS_ISSUED.ANY
Description
Cycles waiting to recover after
Machine Clears or JEClear. Set
Cmask= 1.
Comment
Set Edge to
count
occurrences
Increments each cycle the # of Uops Set Cmask = 1,
issued by the RAT to RS.
Inv = 1to count
stalled cycles
Set Cmask = 1, Inv = 1, Any= 1to
count stalled cycles of this core.
10H
01H
FP_COMP_OPS_EXE.
X87
Counts number of X87 uops
executed.
10H
10H
FP_COMP_OPS_EXE. Counts number of SSE* double
SSE_FP_PACKED_DO precision FP packed uops executed.
UBLE
Vol. 3B A-3
PERFORMANCE-MONITORING EVENTS
Table A-2. Non-Architectural Performance Events In the Processor Core for Intel Core
i7, i5, i3 Processors 2xxx Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
10H
20H
FP_COMP_OPS_EXE. Counts number of SSE* single
SSE_FP_SCALAR_SIN precision FP scalar uops executed.
GLE
10H
40H
FP_COMP_OPS_EXE.
SSE_PACKED SINGLE
10H
80H
FP_COMP_OPS_EXE. Counts number of SSE* double
SSE_SCALAR_DOUBL precision FP scalar uops executed.
E
11H
01H
SIMD_FP_256.PACKE Counts 256-bit packed singleD_SINGLE
precision floating-point instructions
11H
02H
SIMD_FP_256.PACKE Counts 256-bit packed doubleD_DOUBLE
precision floating-point instructions
14H
01H
ARITH.FPU_DIV_ACT Cycles that the divider is active,
IVE
includes INT and FP. Set 'edge =1,
cmask=1' to count the number of
divides.
17H
01H
INSTS_WRITTEN_TO
_IQ.INSTS
24H
01H
L2_RQSTS.DEMAND_ Demand Data Read requests that
DATA_RD_HIT
hit L2 cache
24H
03H
L2_RQSTS.ALL_DEM
AND_DATA_RD
24H
04H
L2_RQSTS.RFO_HITS Counts the number of store RFO
requests that hit the L2 cache.
24H
08H
L2_RQSTS.RFO_MISS Counts the number of store RFO
requests that miss the L2 cache.
24H
0CH
L2_RQSTS.ALL_RFO
Counts all L2 store RFO requests.
24H
10H
L2_RQSTS.CODE_RD
_HIT
Number of instruction fetches that
hit the L2 cache.
24H
20H
L2_RQSTS.CODE_RD
_MISS
Number of instruction fetches that
missed the L2 cache.
24H
30H
L2_RQSTS.ALL_COD
E_RD
Counts all L2 code requests.
24H
40H
L2_RQSTS.PF_HIT
Requests from L2 Hardware
prefetcher that hit L2.
24H
80H
L2_RQSTS.PF_MISS
Requests from L2 Hardware
prefetcher that missed L2.
A-4 Vol. 3B
Description
Counts number of SSE* single
precision FP packed uops executed.
Counts the number of instructions
written into the IQ every cycle.
Counts any demand and L1 HW
prefetch data load requests to L2.
Comment
PERFORMANCE-MONITORING EVENTS
Table A-2. Non-Architectural Performance Events In the Processor Core for Intel Core
i7, i5, i3 Processors 2xxx Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
24H
C0H
L2_RQSTS.ALL_PF
27H
01H
L2_STORE_LOCK_RQ RFOs that miss cache lines
STS.MISS
27H
04H
L2_STORE_LOCK_RQ RFOs that hit cache lines in E state
STS.HIT_E
27H
08H
L2_STORE_LOCK_RQ RFOs that hit cache lines in M state
STS.HIT_M
27H
0FH
L2_STORE_LOCK_RQ RFOs that access cache lines in any
STS.ALL
state
28H
04H
L2_L1D_WB_RQSTS. Not rejected writebacks from L1D
HIT_E
to L2 cache lines in E state.
28H
08H
L2_L1D_WB_RQSTS. Not rejected writebacks from L1D
HIT_M
to L2 cache lines in M state.
2EH
4FH
LONGEST_LAT_CACH This event counts requests
E.REFERENCE
originating from the core that
reference a cache line in the last
level cache.
see Table A-1
2EH
41H
LONGEST_LAT_CACH This event counts each cache miss
E.MISS
condition for references to the last
level cache.
see Table A-1
3CH
00H
CPU_CLK_UNHALTED Counts the number of thread cycles see Table A-1
.THREAD_P
while the thread is not in a halt
state. The thread enters the halt
state when it is running the HLT
instruction. The core frequency may
change from time to time due to
power or thermal throttling.
3CH
01H
CPU_CLK_THREAD_
UNHALTED.REF_XCL
K
Increments at the frequency of
XCLK (100 MHz) when not halted.
48H
01H
L1D_PEND_MISS.PE
NDING
Increments the number of
Counter 2 only;
outstanding L1D misses every cycle. Set Cmask = 1 to
Set Cmaks = 1 and Edge =1 to count count cycles.
occurrences.
49H
01H
DTLB_STORE_MISSE
S.MISS_CAUSES_A_
WALK
Miss in all TLB levels causes an page
walk of any page size
(4K/2M/4M/1G).
Description
Comment
Any requests from L2 Hardware
prefetchers
see Table A-1
Vol. 3B A-5
PERFORMANCE-MONITORING EVENTS
Table A-2. Non-Architectural Performance Events In the Processor Core for Intel Core
i7, i5, i3 Processors 2xxx Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
49H
02H
DTLB_STORE_MISSE Miss in all TLB levels causes a page
S.WALK_COMPLETED walk that completes of any page
size (4K/2M/4M/1G).
49H
04H
DTLB_STORE_MISSE
S.WALK_DURATION
Cycles PMH is busy with this walk.
49H
10H
DTLB_STORE_MISSE
S.STLB_HIT
Store operations that miss the first
TLB level but hit the second and do
not cause page walks
4CH
01H
LOAD_HIT_PRE.SW_
PF
Not SW-prefetch load dispatches
that hit fill buffer allocated for S/W
prefetch.
4CH
02H
LOAD_HIT_PRE.HW_
PF
Not SW-prefetch load dispatches
that hit fill buffer allocated for H/W
prefetch.
4EH
02H
HW_PRE_REQ.DL1_
MISS
Hardware Prefetch requests that
miss the L1D cache. A request is
being counted each time it access
the cache & miss it, including if a
block is applicable or if hit the Fill
Buffer for example.
51H
01H
L1D.REPLACEMENT
Counts the number of lines brought
into the L1 data cache.
51H
02H
L1D.ALLOCATED_IN_ Counts the number of allocations of
M
modified L1D cache lines.
51H
04H
L1D.EVICTION
Counts the number of modified lines
evicted from the L1 data cache due
to replacement.
51H
08H
L1D.ALL_M_REPLAC
EMENT
Cache lines in M state evicted out of
L1D due to Snoop HitM or dirty line
replacement
59H
20H
PARTIAL_RAT_STALL Increments the number of flagsS.FLAGS_MERGE_UO merge uops in flight each cycle.
P
Set Cmask = 1 to count cycles.
59H
40H
PARTIAL_RAT_STALL Cycles with at least one slow LEA
S.SLOW_LEA_WINDO uop allocated.
W
59H
80H
PARTIAL_RAT_STALL Number of Multiply packed/scalar
S.MUL_SINGLE_UOP single precision uops allocated.
A-6 Vol. 3B
Description
Comment
This accounts for
both L1 streamer
and IP-based
(IPP) HW
prefetchers.
PERFORMANCE-MONITORING EVENTS
Table A-2. Non-Architectural Performance Events In the Processor Core for Intel Core
i7, i5, i3 Processors 2xxx Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
5BH
0CH
RESOURCE_STALLS2. Cycles stalled due to free list empty
ALL_FL_EMPTY
5BH
0FH
RESOURCE_STALLS2. Cycles stalled due to control
ALL_PRF_CONTROL structures full for physical registers
5BH
40H
RESOURCE_STALLS2. Cycles Allocator is stalled due
BOB_FULL
Branch Order Buffer.
5BH
4FH
RESOURCE_STALLS2. Cycles stalled due to out of order
OOO_RSRC
resources full
5CH
01H
CPL_CYCLES.RING0
Unhalted core cycles when the
thread is in ring 0
5CH
02H
CPL_CYCLES.RING12
3
Unhalted core cycles when the
thread is not in ring 0
5EH
01H
RS_EVENTS.EMPTY_ Cycles the RS is empty for the
CYCLES
thread.
60H
01H
OFFCORE_REQUEST
S_OUTSTANDING.DE
MAND_DATA_RD
Offcore outstanding Demand Data
Read transactions in SQ to uncore.
Set Cmask=1 to count cycles.
60H
04H
OFFCORE_REQUEST
S_OUTSTANDING.DE
MAND_RFO
Offcore outstanding RFO store
transactions in SQ to uncore. Set
Cmask=1 to count cycles.
60H
08H
OFFCORE_REQUEST
S_OUTSTANDING.AL
L_DATA_RD
Offcore outstanding cacheable data
read transactions in SQ to uncore.
Set Cmask=1 to count cycles.
63H
01H
LOCK_CYCLES.SPLIT_ Cycles in which the L1D and L2 are
LOCK_UC_LOCK_DUR locked, due to a UC lock or split lock.
ATION
63H
02H
LOCK_CYCLES.CACHE Cycles in which the L1D is locked.
_LOCK_DURATION
79H
02H
IDQ.EMPTY
Counts cycles the IDQ is empty.
79H
04H
IDQ.MITE_UOPS
Increment each cycle # of uops
delivered to IDQ from MITE path.
79H
08H
IDQ.DSB_UOPS
Increment each cycle. # of uops
delivered to IDQ from DSB path.
Description
Set Cmask = 1 to count cycles.
Set Cmask = 1 to count cycles.
Comment
Use Edge to
count transition
Can combine
Umask 04H and
20H
Can combine
Umask 08H and
10H
Vol. 3B A-7
PERFORMANCE-MONITORING EVENTS
Table A-2. Non-Architectural Performance Events In the Processor Core for Intel Core
i7, i5, i3 Processors 2xxx Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
Description
Comment
79H
10H
IDQ.MS_DSB_UOPS
Increment each cycle # of uops
delivered to IDQ when MS busy by
DSB. Set Cmask = 1 to count cycles
MS is busy. Set Cmask=1 and Edge
=1 to count MS activations.
Can combine
Umask 08H and
10H
79H
20H
IDQ.MS_MITE_UOPS
Increment each cycle # of uops
Can combine
delivered to IDQ when MS is busy by Umask 04H and
MITE. Set Cmask = 1 to count cycles. 20H
79H
30H
IDQ.MS_UOPS
Increment each cycle # of uops
Can combine
delivered to IDQ from MS by either Umask 04H, 08H
DSB or MITE. Set Cmask = 1 to count and 30H
cycles.
80H
02H
ICACHE.MISSES
Number of Instruction Cache,
Streaming Buffer and Victim Cache
Misses. Includes UC accesses.
85H
01H
ITLB_MISSES.MISS_C Misses in all ITLB levels that cause
AUSES_A_WALK
page walks
85H
02H
ITLB_MISSES.WALK_
COMPLETED
Misses in all ITLB levels that cause
completed page walks
85H
04H
ITLB_MISSES.WALK_
DURATION
Cycle PMH is busy with a walk.
85H
10H
ITLB_MISSES.STLB_H Number of cache load STLB hits. No
IT
page walk.
87H
01H
ILD_STALL.LCP
Stalls caused by changing prefix
length of the instruction.
87H
04H
ILD_STALL.IQ_FULL
Stall cycles due to IQ is full.
88H
01H
BR_INST_EXEC.COND Qualify conditional near branch
instructions executed, but not
necessarily retired.
88H
02H
BR_INST_EXEC.DIRE
CT_JMP
88H
04H
BR_INST_EXEC.INDIR Qualify executed indirect near
ECT_JMP_NON_CALL branch instructions that are not
_RET
calls nor returns.
Must combine
with umask 80H
88H
08H
BR_INST_EXEC.RETU Qualify indirect near branches that
RN_NEAR
have a return mnemonic.
Must combine
with umask 80H
A-8 Vol. 3B
Must combine
with umask 40H,
80H
Qualify all unconditional near branch Must combine
instructions excluding calls and
with umask 80H
indirect branches.
PERFORMANCE-MONITORING EVENTS
Table A-2. Non-Architectural Performance Events In the Processor Core for Intel Core
i7, i5, i3 Processors 2xxx Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
88H
10H
BR_INST_EXEC.DIRE
CT_NEAR_CALL
88H
20H
BR_INST_EXEC.INDIR Qualify indirect near calls, including
ECT_NEAR_CALL
both register and memory indirect,
executed.
Must combine
with umask 80H
88H
40H
BR_INST_EXEC.NON
TAKEN
Applicable to
umask 01H only
88H
80H
BR_INST_EXEC.TAKE Qualify taken near branches
N
executed. Must combine with
01H,02H, 04H, 08H, 10H, 20H
88H
FFH
BR_INST_EXEC.ALL_ Counts all near executed branches
BRANCHES
(not necessarily retired).
89H
01H
BR_MISP_EXEC.CON
D
89H
04H
BR_MISP_EXEC.INDIR Qualify mispredicted indirect near
ECT_JMP_NON_CALL branch instructions that are not
_RET
calls nor returns.
Must combine
with umask 80H
89H
08H
BR_MISP_EXEC.RETU Qualify mispredicted indirect near
RN_NEAR
branches that have a return
mnemonic.
Must combine
with umask 80H
89H
10H
BR_MISP_EXEC.DIRE
CT_NEAR_CALL
89H
20H
BR_MISP_EXEC.INDIR Qualify mispredicted indirect near
ECT_NEAR_CALL
calls, including both register and
memory indirect, executed.
Must combine
with umask 80H
89H
40H
BR_MISP_EXEC.NON
TAKEN
Applicable to
umask 01H only
89H
80H
BR_MISP_EXEC.TAKE Qualify mispredicted taken near
N
branches executed. Must combine
with 01H,02H, 04H, 08H, 10H, 20H
89H
FFH
BR_MISP_EXEC.ALL_ Counts all near executed branches
BRANCHES
(not necessarily retired).
9CH
01H
IDQ_UOPS_NOT_DEL Count number of non-delivered
IVERED.CORE
uops to RAT per thread.
Description
Comment
Qualify unconditional near call
branch instructions, excluding non
call branch, executed.
Must combine
with umask 80H
Qualify non-taken near branches
executed.
Qualify conditional near branch
instructions mispredicted.
Must combine
with umask 40H,
80H
Qualify mispredicted unconditional Must combine
near call branch instructions,
with umask 80H
excluding non call branch, executed.
Qualify mispredicted non-taken
near branches executed,.
Use Cmask to
qualify uop b/w
Vol. 3B A-9
PERFORMANCE-MONITORING EVENTS
Table A-2. Non-Architectural Performance Events In the Processor Core for Intel Core
i7, i5, i3 Processors 2xxx Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
Description
A1H
01H
UOPS_DISPATCHED_
PORT.PORT_0
Cycles which a Uop is dispatched on
port 0.
A1H
02H
UOPS_DISPATCHED_ Cycles which a Uop is dispatched on
PORT.PORT_1
port 1.
A1H
04H
UOPS_DISPATCHED_ Cycles which a load uop is
PORT.PORT_2_LD
dispatched on port 2.
A1H
08H
UOPS_DISPATCHED_ Cycles which a store address uop is
PORT.PORT_2_STA
dispatched on port 2.
A1H
0CH
UOPS_DISPATCHED_ Cycles which a Uop is dispatched on
PORT.PORT_2
port 2.
A1H
10H
UOPS_DISPATCHED_ Cycles which a load uop is
PORT.PORT_3_LD
dispatched on port 3.
A1H
20H
UOPS_DISPATCHED_ Cycles which a store address uop is
PORT.PORT_3_STA
dispatched on port 3.
A1H
30H
UOPS_DISPATCHED_ Cycles which a Uop is dispatched on
PORT.PORT_3
port 3.
A1H
40H
UOPS_DISPATCHED_
PORT.PORT_4
Cycles which a Uop is dispatched on
port 4.
A1H
80H
UOPS_DISPATCHED_
PORT.PORT_5
Cycles which a Uop is dispatched on
port 5.
A2H
01H
RESOURCE_STALLS.
ANY
Cycles Allocation is stalled due to
Resource Related reason.
A2H
02H
RESOURCE_STALLS.L Counts the cycles of stall due to lack
B
of load buffers.
A2H
04H
RESOURCE_STALLS.R Cycles stalled due to no eligible RS
S
entry available.
A2H
08H
RESOURCE_STALLS.S Cycles stalled due to no store
B
buffers available. (not including
draining form sync).
A2H
10H
RESOURCE_STALLS.R Cycles stalled due to re-order buffer
OB
full.
A2H
20H
RESOURCE_STALLS.F Cycles stalled due to writing the
CSW
FPU control word.
A2H
40H
RESOURCE_STALLS.
MXCSR
A-10 Vol. 3B
Cycles stalled due to the MXCSR
register rename occurring to close
to a previous MXCSR rename.
Comment
PERFORMANCE-MONITORING EVENTS
Table A-2. Non-Architectural Performance Events In the Processor Core for Intel Core
i7, i5, i3 Processors 2xxx Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
Description
A2H
80H
RESOURCE_STALLS.
OTHER
Cycles stalled while execution was
stalled due to other resource issues.
ABH
01H
DSB2MITE_SWITCHE
S.COUNT
Number of DSB to MITE switches.
ABH
02H
DSB2MITE_SWITCHE
S.PENALTY_CYCLES
Cycles DSB to MITE switches caused
delay.
ACH
02H
DSB_FILL.OTHER_CA Cases of cancelling valid DSB fill not
NCEL
because of exceeding way limit
ACH
08H
DSB_FILL.EXCEED_D
SB_LINES
DSB Fill encountered > 3 DSB lines.
ACH
0AH
DSB_FILL.ALL_CANC
EL
Cases of cancelling valid Decode
Stream Buffer (DSB) fill not because
of exceeding way limit
AEH
01H
ITLB.ITLB_FLUSH
Counts the number of ITLB flushes,
includes 4k/2M/4M pages.
B0H
01H
OFFCORE_REQUEST Demand data read requests sent to
S.DEMAND_DATA_RD uncore.
B0H
04H
OFFCORE_REQUEST
S.DEMAND_RFO
Demand RFO read requests sent to
uncore., including regular RFOs,
locks, ItoM
B0H
08H
OFFCORE_REQUEST
S.ALL_DATA_RD
Data read requests sent to uncore
(demand and prefetch).
B1H
01H
UOPS_DISPATCHED.T Counts total number of uops to be
HREAD
dispatched per-thread each cycle.
Set Cmask = 1, INV =1 to count stall
cycles.
B1H
02H
UOPS_DISPATCHED.C Counts total number of uops to be
ORE
dispatched per-core each cycle.
B2H
01H
OFFCORE_REQUEST
S_BUFFER.SQ_FULL
B6H
01H
AGU_BYPASS_CANCE Counts executed load operations
L.COUNT
with all the following traits: 1.
addressing of the format [base +
offset], 2. the offset is between 1
and 2047, 3. the address specified
in the base register is in one page
and the address [base+offset] is in
another page.
Comment
Do not need to
set ANY
Offcore requests buffer cannot take
more entries for this thread core.
Vol. 3B A-11
PERFORMANCE-MONITORING EVENTS
Table A-2. Non-Architectural Performance Events In the Processor Core for Intel Core
i7, i5, i3 Processors 2xxx Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
B7H
01H
OFF_CORE_RESPONS see Section 30.8.5, “Off-core
Requires
E_0
Response Performance Monitoring”; programming
PMC0 only.
MSR 01A6H
BBH
01H
OFF_CORE_RESPONS See Section 30.8.5, “Off-core
Requires
E_1
Response Performance Monitoring”. programming
PMC3 only.
MSR 01A7H
BDH
01H
TLB_FLUSH.DTLB_T
HREAD
DTLB flush attempts of the threadspecific entries
BDH
20H
TLB_FLUSH.STLB_A
NY
Count number of STLB flush
attempts
BFH
05H
L1D_BLOCKS.BANK_
CONFLICT_CYCLES
Cycles when dispatched loads are
cancelled due to L1D bank conflicts
with other load ports
C0H
00H
INST_RETIRED.ANY_ Number of instructions at
P
retirement
See Table A-1
C0H
01H
INST_RETIRED.PREC
_DIST
Precise instruction retired event
with HW to reduce effect of PEBS
shadow in IP distribution
PMC1 only; Must
quiesce other
PMCs.
C0H
02H
INST_RETIRED.X87
X87 instruction retired event
C1H
02H
OTHER_ASSISTS.ITL
B_MISS_RETIRED
Instructions that experienced an
ITLB miss.
C1H
08H
OTHER_ASSISTS.AVX Number of assists associated with
_STORE
256-bit AVX store operations.
C1H
10H
OTHER_ASSISTS.AVX Number of transitions from AVX_TO_SSE
256 to legacy SSE when penalty
applicable.
C1H
20H
OTHER_ASSISTS.SSE Number of transitions from SSE to
_TO_AVX
AVX-256 when penalty applicable.
C2H
01H
UOPS_RETIRED.ALL
Counts the number of micro-ops
Supports PEBS
retired, Use cmask=1 and invert to
count active cycles or stalled cycles.
C2H
02H
UOPS_RETIRED.RETI
RE_SLOTS
Counts the number of retirement
slots used each cycle.
C3H
02H
MACHINE_CLEARS.M
EMORY_ORDERING
Counts the number of machine
clears due to memory order
conflicts.
A-12 Vol. 3B
Description
Comment
cmask=1
PERFORMANCE-MONITORING EVENTS
Table A-2. Non-Architectural Performance Events In the Processor Core for Intel Core
i7, i5, i3 Processors 2xxx Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
Description
C3H
04H
MACHINE_CLEARS.S
MC
Counts the number of times that a
program writes to a code section.
C3H
20H
MACHINE_CLEARS.M
ASKMOV
Counts the number of executed
AVX masked load operations that
refer to an illegal address range
with the mask bits set to 0.
C4H
00H
BR_INST_RETIRED.A
LL_BRANCHES
Branch instructions at retirement
See Table A-1
C4H
01H
BR_INST_RETIRED.C
ONDITIONAL
Counts the number of conditional
branch instructions retired.
Supports PEBS
C4H
02H
BR_INST_RETIRED.N Direct and indirect near call
EAR_CALL
instructions retired.
C4H
04H
BR_INST_RETIRED.A
LL_BRANCHES
C4H
08H
BR_INST_RETIRED.N Counts the number of near return
EAR_RETURN
instructions retired.
C4H
10H
BR_INST_RETIRED.N Counts the number of not taken
OT_TAKEN
branch instructions retired.
C4H
20H
BR_INST_RETIRED.N Number of near taken branches
EAR_TAKEN
retired.
C4H
40H
BR_INST_RETIRED.F
AR_BRANCH
C5H
00H
BR_MISP_RETIRED.A Mispredicted branch instructions at
LL_BRANCHES
retirement
See Table A-1
C5H
01H
BR_MISP_RETIRED.C
ONDITIONAL
Supports PEBS
C5H
02H
BR_MISP_RETIRED.N Direct and indirect mispredicted
EAR_CALL
near call instructions retired.
C5H
04H
BR_MISP_RETIRED.A Mispredicted macro branch
LL_BRANCHES
instructions retired.
C5H
10H
BR_MISP_RETIRED.N Mispredicted not taken branch
OT_TAKEN
instructions retired.
C5H
20H
BR_MISP_RETIRED.T Mispredicted taken branch
AKEN
instructions retired.
CAH
02H
FP_ASSIST.X87_OUT Number of X87 assists due to
PUT
output value.
Comment
Counts the number of branch
instructions retired.
Number of far branches retired.
Mispredicted conditional branch
instructions retired.
Vol. 3B A-13
PERFORMANCE-MONITORING EVENTS
Table A-2. Non-Architectural Performance Events In the Processor Core for Intel Core
i7, i5, i3 Processors 2xxx Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
Description
CAH
04H
FP_ASSIST.X87_INP
UT
Number of X87 assists due to input
value.
CAH
08H
FP_ASSIST.SIMD_OU
TPUT
Number of SIMD FP assists due to
Output values
CAH
10H
FP_ASSIST.SIMD_INP Number of SIMD FP assists due to
UT
input values
CAH
1EH
FP_ASSIST.ANY
CCH
20H
ROB_MISC_EVENTS.L Count cases of saving new LBR
BR_INSERTS
records by hardware.
CDH
01H
MEM_TRANS_RETIR
ED.LOAD_LATENCY
Sample loads with specified latency Specify threshold
threshold. PMC3 only.
in MSR 0x3F6
CDH
02H
MEM_TRANS_RETIR
ED.PRECISE_STORE
Sample stores and collect precise
store operation via PEBS record.
PMC3 only.
D0H
01H
MEM_UOP_RETIRED. Qualify retired memory uops that
Supports PEBS
LOADS
are loads. Combine with umask 10H,
20H, 40H, 80H.
D0H
02H
MEM_UOP_RETIRED. Qualify retired memory uops that
STORES
are stores. Combine with umask
10H, 20H, 40H, 80H.
D0H
10H
MEM_UOP_RETIRED. Qualify retired memory uops with
STLB_MISS
STLB miss. Must combine with
umask 01H, 02H, to produce counts.
D0H
20H
MEM_UOP_RETIRED. Qualify retired memory uops with
LOCK
lock. Must combine with umask 01H,
02H, to produce counts.
D0H
40H
MEM_UOP_RETIRED. Qualify retired memory uops with
SPLIT
line split. Must combine with umask
01H, 02H, to produce counts.
D0H
80H
MEM_UOP_RETIRED. Qualify any retired memory uops.
ALL
Must combine with umask 01H,
02H, to produce counts.
D1H
01H
MEM_LOAD_UOPS_R Retired load uops with L1 cache hits Supports PEBS
ETIRED.L1_HIT
as data sources.
D1H
02H
MEM_LOAD_UOPS_R Retired load uops with L2 cache hits
ETIRED.L2_HIT
as data sources.
A-14 Vol. 3B
Comment
Cycles with any input/output SSE*
or FP assists
See Section
30.8.4.3
PERFORMANCE-MONITORING EVENTS
Table A-2. Non-Architectural Performance Events In the Processor Core for Intel Core
i7, i5, i3 Processors 2xxx Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
D1H
04H
MEM_LOAD_UOPS_R Retired load uops which data
ETIRED.LLC_HIT
sources were data hits in LLC
without snoops required.
D1H
40H
MEM_LOAD_UOPS_R Retired load uops which data
ETIRED.HIT_LFB
sources were load uops missed L1
but hit FB due to preceding miss to
the same cache line with data not
ready.
D2H
01H
MEM_LOAD_UOPS_L
LC_HIT_RETIRED.XS
NP_MISS
Retired load uops which data
Supports PEBS
sources were LLC hit and cross-core
snoop missed in on-pkg core cache.
D2H
02H
MEM_LOAD_UOPS_L
LC_HIT_RETIRED.XS
NP_HIT
Retired load uops which data
sources were LLC and cross-core
snoop hits in on-pkg core cache.
D2H
04H
MEM_LOAD_UOPS_L
LC_HIT_RETIRED.XS
NP_HITM
Retired load uops which data
sources were HitM responses from
shared LLC.
D2H
08H
MEM_LOAD_UOPS_L
LC_HIT_RETIRED.XS
NP_NONE
Retired load uops which data
sources were hits in LLC without
snoops required.
D4H
02H
MEM_LOAD_UOPS_M Retired load uops with unknown
ISC_RETIRED.LLC_MI information as data source in cache
SS
serviced the load.
F0H
01H
L2_TRANS.DEMAND_ Demand Data Read requests that
DATA_RD
access L2 cache
F0H
02H
L2_TRANS.RFO
F0H
04H
L2_TRANS.CODE_RD L2 cache accesses when fetching
instructions
F0H
08H
L2_TRANS.ALL_PF
L2 or LLC HW prefetches that
access L2 cache
F0H
10H
L2_TRANS.L1D_WB
L1D writebacks that access L2
cache
F0H
20H
L2_TRANS.L2_FILL
L2 fill requests that access L2 cache
F0H
40H
L2_TRANS.L2_WB
L2 writebacks that access L2 cache
F0H
80H
L2_TRANS.ALL_REQ
UESTS
Transactions accessing L2 pipe
Description
Comment
Supports PEBS.
RFO requests that access L2 cache
including rejects.
Vol. 3B A-15
PERFORMANCE-MONITORING EVENTS
Table A-2. Non-Architectural Performance Events In the Processor Core for Intel Core
i7, i5, i3 Processors 2xxx Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
Description
Comment
F1H
01H
L2_LINES_IN.I
L2 cache lines in I state filling L2
Counting does
not cover rejects.
F1H
02H
L2_LINES_IN.S
L2 cache lines in S state filling L2
Counting does
not cover rejects.
F1H
04H
L2_LINES_IN.E
L2 cache lines in E state filling L2
Counting does
not cover rejects.
F1H
07H
L2_LINES_IN.ALL
L2 cache lines filling L2
Counting does
not cover rejects.
F2H
01H
L2_LINES_OUT.DEMA Clean L2 cache lines evicted by
ND_CLEAN
demand
F2H
02H
L2_LINES_OUT.DEMA Dirty L2 cache lines evicted by
ND_DIRTY
demand
F2H
04H
L2_LINES_OUT.PF_C
LEAN
F2H
08H
L2_LINES_OUT.PF_DI Dirty L2 cache lines evicted by L2
RTY
prefetch
F2H
0AH
L2_LINES_OUT.DIRT
Y_ALL
F4H
10H
SQ_MISC.SPLIT_LOCK Split locks in SQ
Clean L2 cache lines evicted by L2
prefetch
Dirty L2 cache lines filling the L2
Counting does
not cover rejects.
Non-architectural Performance monitoring events that are located in the uncore subsystem are implementation specific between different platforms using processors
based on Intel microarchitecture Sandy Bridge. Processors with CPUID signature of
DisplayFamily_DisplayModel 06_2AH support performance events listed in Table A-3.
A-16 Vol. 3B
PERFORMANCE-MONITORING EVENTS
Table A-3. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7, i5, i3 Processor 2xxx Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
22H
01H
UNC_CBO_XSNP_RE
SPONSE.RSPIHITI
22H
02H
UNC_CBO_XSNP_RE
SPONSE.RSPIHITFSE
22H
04H
UNC_CBO_XSNP_RE
SPONSE.RSPSHITFSE
22H
08H
UNC_CBO_XSNP_RE
SPONSE.RSPSFWDM
22H
01H
UNC_CBO_XSNP_RE
SPONSE.RSPIFWDM
22H
20H
UNC_CBO_XSNP_RE Filter on cross-core snoops resulted in
SPONSE.AND_EXTER external snoop request. Must combine
with at least one of 01H, 02H, 04H,
NAL
08H, 10H
22H
40H
UNC_CBO_XSNP_RE Filter on cross-core snoops resulted in
SPONSE.AND_XCORE core request. Must combine with at
least one of 01H, 02H, 04H, 08H, 10H
22H
80H
UNC_CBO_XSNP_RE Filter on cross-core snoops resulted in
SPONSE.AND_XCORE LLC evictions. Must combine with at
least one of 01H, 02H, 04H, 08H, 10H
34H
01H
UNC_CBO_CACHE_LO LLC lookup request that access cache
OKUP.M
and found line in M-state.
34H
02H
UNC_CBO_CACHE_LO LLC lookup request that access cache
OKUP.E
and found line in E-state.
34H
04H
UNC_CBO_CACHE_LO LLC lookup request that access cache
OKUP.S
and found line in S-state.
34H
08H
UNC_CBO_CACHE_LO LLC lookup request that access cache
OKUP.I
and found line in I-state.
34H
10H
UNC_CBO_CACHE_LO Filter on processor core initiated
OKUP.AND_READ
cacheable read requests. Must
combine with at least one of 01H,
02H, 04H, 08H
34H
20H
UNC_CBO_CACHE_LO Filter on processor core initiated
OKUP.AND_READ
cacheable write requests. Must
combine with at least one of 01H,
02H, 04H, 08H
Description
Comment
Snoop responses received from
processor cores to requests initiated
by this Cbox.
Must combine
with one of the
umask values
of 20H, 40H,
80H
Must combine
with one of the
umask values
of 10H, 20H,
40H, 80H
Vol. 3B A-17
PERFORMANCE-MONITORING EVENTS
Table A-3. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7, i5, i3 Processor 2xxx Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
34H
40H
UNC_CBO_CACHE_LO Filter on external snoop requests.
OKUP.AND_EXTSNP Must combine with at least one of
01H, 02H, 04H, 08H
34H
80H
UNC_CBO_CACHE_LO Filter on any IRQ or IPQ initiated
OKUP.AND_ANY
requests including uncacheable, noncoherent requests. Must combine with
at least one of 01H, 02H, 04H, 08H
80H
01H
UNC_IMPH_CBO_TRK Counts cycles weighted by the
Counter 0 only
_OCCUPANCY.ALL
number of core-outgoing valid entries.
Valid entries are between allocation
to the first of IDIO or DRSO messages.
Accounts for coherent and incoherent traffic
81H
01H
UNC_IMPH_CBO_TRK Counts the number of core-outgoing
_REQUEST.ALL
entries. Accounts for coherent and incoherent traffic
81H
20H
UNC_IMPH_CBO_TRK Counts the number of allocated write
_REQUEST.WRITES
entries, include full, partial, and
evictions.
81H
80H
UNC_IMPH_CBO_TRK Counts the number of evictions
_REQUEST.EVICTION allocated.
S
83H
01H
UNC_IMPH_COH_TR
K_OCCUPANCY.ALL
Counts cycles weighted by the
Counter 0 only
number of core-outgoing valid entries
in the coherent tracker queue.
84H
01H
UNC_IMPH_COH_TR
K_REQUEST.ALL
Counts the number of core-outgoing
entries in the coherent tracker queue.
A.3
Description
Comment
PERFORMANCE MONITORING EVENTS FOR
INTEL® CORE™I7 PROCESSOR FAMILY AND XEON
PROCESSOR FAMILY
Processors based on the Intel microarchitecture code name Nehalem support the
architectural and non-architectural performance-monitoring events listed in Table
A-1 and Table A-4. The events in Table A-4 generally applies to processors with
A-18 Vol. 3B
PERFORMANCE-MONITORING EVENTS
CPUID signature of DisplayFamily_DisplayModel encoding with the following values:
06_1AH, 06_1EH, 06_1FH, and 06_2EH. However, Intel Xeon processors with CPUID
signature of DisplayFamily_DisplayModel 06_2EH have a small number of events that
are not supported in processors with CPUID signature 06_1AH, 06_1EH, and
06_1FH. These events are noted in the comment column.
In addition, these processors (CPUID signature of DisplayFamily_DisplayModel
06_1AH, 06_1EH, 06_1FH) also support the following non-architectural, productspecific uncore performance-monitoring events listed in Table A-5.
Fixed counters in the core PMU support the architecture events defined in Table A-9.
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
04H
07H
SB_DRAIN.ANY
Counts the number of store buffer
drains.
06H
04H
STORE_BLOCKS.AT_
RET
Counts number of loads delayed
with at-Retirement block code. The
following loads need to be executed
at retirement and wait for all senior
stores on the same thread to be
drained: load splitting across 4K
boundary (page split), load
accessing uncacheable (UC or
USWC) memory, load lock, and load
with page table in UC or USWC
memory region.
06H
08H
STORE_BLOCKS.L1D
_BLOCK
Cacheable loads delayed with L1D
block code.
07H
01H
PARTIAL_ADDRESS_
ALIAS
Counts false dependency due to
partial address aliasing.
08H
01H
DTLB_LOAD_MISSES. Counts all load misses that cause a
ANY
page walk.
08H
02H
DTLB_LOAD_MISSES. Counts number of completed page
WALK_COMPLETED
walks due to load miss in the STLB.
08H
10H
DTLB_LOAD_MISSES. Number of cache load STLB hits.
STLB_HIT
08H
20H
DTLB_LOAD_MISSES. Number of DTLB cache load misses
PDE_MISS
where the low part of the linear to
physical address translation was
missed.
Description
Comment
Vol. 3B A-19
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
08H
80H
DTLB_LOAD_MISSES. Counts number of completed large
LARGE_WALK_COMP page walks due to load miss in the
LETED
STLB.
0BH
01H
MEM_INST_RETIRED. Counts the number of instructions
LOADS
with an architecturally-visible load
retired on the architected path.
0BH
02H
MEM_INST_RETIRED. Counts the number of instructions
STORES
with an architecturally-visible store
retired on the architected path.
0BH
10H
MEM_INST_RETIRED. Counts the number of instructions
LATENCY_ABOVE_T exceeding the latency specified
with ld_lat facility.
HRESHOLD
0CH
01H
MEM_STORE_RETIRE The event counts the number of
D.DTLB_MISS
retired stores that missed the DTLB.
The DTLB miss is not counted if the
store operation causes a fault. Does
not counter prefetches. Counts both
primary and secondary misses to
the TLB.
0EH
01H
UOPS_ISSUED.ANY
0EH
01H
UOPS_ISSUED.STALL Counts the number of cycles no
ED_CYCLES
Uops issued by the Register
Allocation Table to the Reservation
Station, i.e. the UOPs issued from
the front end to the back end.
0EH
02H
UOPS_ISSUED.FUSED Counts the number of fused Uops
that were issued from the Register
Allocation Table to the Reservation
Station.
0FH
01H
MEM_UNCORE_RETI Counts number of memory load
RED.L3_DATA_MISS_ instructions retired where the
UNKNOWN
memory reference missed L3 and
data source is unknown.
A-20 Vol. 3B
Description
Comment
In conjunction
with ld_lat
facility
Counts the number of Uops issued
by the Register Allocation Table to
the Reservation Station, i.e. the
UOPs issued from the front end to
the back end.
set “invert=1,
cmask = 1“
Available only for
CPUID signature
06_2EH
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
0FH
02H
MEM_UNCORE_RETI
RED.OTHER_CORE_L
2_HITM
0FH
08H
MEM_UNCORE_RETI Counts number of memory load
RED.REMOTE_CACHE instructions retired where the
_LOCAL_HOME_HIT memory reference missed the L1,
L2 and L3 caches and HIT in a
remote socket's cache. Only counts
locally homed lines.
0FH
10H
MEM_UNCORE_RETI
RED.REMOTE_DRAM
Counts number of memory load
instructions retired where the
memory reference missed the L1,
L2 and L3 caches and was remotely
homed. This includes both DRAM
access and HITM in a remote
socket's cache for remotely homed
lines.
0FH
20H
MEM_UNCORE_RETI
RED.LOCAL_DRAM
Counts number of memory load
instructions retired where the
memory reference missed the L1,
L2 and L3 caches and required a
local socket memory reference. This
includes locally homed cachelines
that were in a modified state in
another socket.
0FH
80H
MEM_UNCORE_RETI
RED.UNCACHEABLE
Counts number of memory load
instructions retired where the
memory reference missed the L1,
L2 and L3 caches and to perform
I/O.
10H
01H
FP_COMP_OPS_EXE.
X87
Counts the number of FP
Computational Uops Executed. The
number of FADD, FSUB, FCOM,
FMULs, integer MULsand IMULs,
FDIVs, FPREMs, FSQRTS, integer
DIVs, and IDIVs. This event does not
distinguish an FADD used in the
middle of a transcendental flow
from a separate FADD instruction.
Description
Comment
Counts number of memory load
instructions retired where the
memory reference hit modified data
in a sibling core residing on the
same socket.
Available only for
CPUID signature
06_2EH
Vol. 3B A-21
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
Description
10H
02H
FP_COMP_OPS_EXE.
MMX
Counts number of MMX Uops
executed.
10H
04H
FP_COMP_OPS_EXE.
SSE_FP
Counts number of SSE and SSE2 FP
uops executed.
10H
08H
FP_COMP_OPS_EXE.
SSE2_INTEGER
Counts number of SSE2 integer
uops executed.
10H
10H
FP_COMP_OPS_EXE.
SSE_FP_PACKED
Counts number of SSE FP packed
uops executed.
10H
20H
FP_COMP_OPS_EXE.
SSE_FP_SCALAR
Counts number of SSE FP scalar
uops executed.
10H
40H
FP_COMP_OPS_EXE. Counts number of SSE* FP single
SSE_SINGLE_PRECISI precision uops executed.
ON
10H
80H
FP_COMP_OPS_EXE.
SSE_DOUBLE_PRECI
SION
12H
01H
SIMD_INT_128.PACK Counts number of 128 bit SIMD
ED_MPY
integer multiply operations.
12H
02H
SIMD_INT_128.PACK Counts number of 128 bit SIMD
ED_SHIFT
integer shift operations.
12H
04H
SIMD_INT_128.PACK Counts number of 128 bit SIMD
integer pack operations.
12H
08H
SIMD_INT_128.UNPA Counts number of 128 bit SIMD
CK
integer unpack operations.
12H
10H
SIMD_INT_128.PACK Counts number of 128 bit SIMD
ED_LOGICAL
integer logical operations.
12H
20H
SIMD_INT_128.PACK Counts number of 128 bit SIMD
ED_ARITH
integer arithmetic operations.
12H
40H
SIMD_INT_128.SHUF Counts number of 128 bit SIMD
FLE_MOVE
integer shuffle and move
operations.
13H
01H
LOAD_DISPATCH.RS
A-22 Vol. 3B
Counts number of SSE* FP double
precision uops executed.
Counts number of loads dispatched
from the Reservation Station that
bypass the Memory Order Buffer.
Comment
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
13H
02H
LOAD_DISPATCH.RS_ Counts the number of delayed RS
DELAYED
dispatches at the stage latch. If an
RS dispatch can not bypass to LB, it
has another chance to dispatch
from the one-cycle delayed staging
latch before it is written into the LB.
13H
04H
LOAD_DISPATCH.MO
B
13H
07H
LOAD_DISPATCH.ANY Counts all loads dispatched from the
Reservation Station.
14H
01H
ARITH.CYCLES_DIV_
BUSY
Description
Comment
Counts the number of loads
dispatched from the Reservation
Station to the Memory Order Buffer.
Counts the number of cycles the
Count may be
divider is busy executing divide or
incorrect When
square root operations. The divide SMT is on.
can be integer, X87 or Streaming
SIMD Extensions (SSE). The square
root operation can be either X87 or
SSE.
Set 'edge =1, invert=1, cmask=1' to
count the number of divides.
14H
02H
ARITH.MUL
Counts the number of multiply
operations executed. This includes
integer as well as floating point
multiply operations but excludes
DPPS mul and MPSAD.
17H
01H
INST_QUEUE_WRITE Counts the number of instructions
S
written into the instruction queue
every cycle.
18H
01H
INST_DECODED.DEC0 Counts number of instructions that
require decoder 0 to be decoded.
Usually, this means that the
instruction maps to more than 1
uop.
19H
01H
TWO_UOP_INSTS_D
ECODED
Count may be
incorrect When
SMT is on
An instruction that generates two
uops was decoded.
Vol. 3B A-23
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
Description
Comment
1EH
01H
INST_QUEUE_WRITE
_CYCLES
This event counts the number of
cycles during which instructions are
written to the instruction queue.
Dividing this counter by the number
of instructions written to the
instruction queue
(INST_QUEUE_WRITES) yields the
average number of instructions
decoded each cycle. If this number is
less than four and the pipe stalls,
this indicates that the decoder is
failing to decode enough
instructions per cycle to sustain the
4-wide pipeline.
If SSE*
instructions that
are 6 bytes or
longer arrive one
after another,
then front end
throughput may
limit execution
speed. In such
case,
20H
01H
LSD_OVERFLOW
Counts number of loops that can’t
stream from the instruction queue.
24H
01H
L2_RQSTS.LD_HIT
Counts number of loads that hit the
L2 cache. L2 loads include both L1D
demand misses as well as L1D
prefetches. L2 loads can be
rejected for various reasons. Only
non rejected loads are counted.
24H
02H
L2_RQSTS.LD_MISS
Counts the number of loads that
miss the L2 cache. L2 loads include
both L1D demand misses as well as
L1D prefetches.
24H
03H
L2_RQSTS.LOADS
Counts all L2 load requests. L2 loads
include both L1D demand misses as
well as L1D prefetches.
24H
04H
L2_RQSTS.RFO_HIT
Counts the number of store RFO
requests that hit the L2 cache. L2
RFO requests include both L1D
demand RFO misses as well as L1D
RFO prefetches. Count includes WC
memory requests, where the data is
not fetched but the permission to
write the line is required.
A-24 Vol. 3B
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
24H
08H
L2_RQSTS.RFO_MISS Counts the number of store RFO
requests that miss the L2 cache. L2
RFO requests include both L1D
demand RFO misses as well as L1D
RFO prefetches.
24H
0CH
L2_RQSTS.RFOS
24H
10H
L2_RQSTS.IFETCH_H Counts number of instruction
IT
fetches that hit the L2 cache. L2
instruction fetches include both L1I
demand misses as well as L1I
instruction prefetches.
24H
20H
L2_RQSTS.IFETCH_M Counts number of instruction
ISS
fetches that miss the L2 cache. L2
instruction fetches include both L1I
demand misses as well as L1I
instruction prefetches.
24H
30H
L2_RQSTS.IFETCHES Counts all instruction fetches. L2
instruction fetches include both L1I
demand misses as well as L1I
instruction prefetches.
24H
40H
L2_RQSTS.PREFETC
H_HIT
Counts L2 prefetch hits for both
code and data.
24H
80H
L2_RQSTS.PREFETC
H_MISS
Counts L2 prefetch misses for both
code and data.
24H
C0H
L2_RQSTS.PREFETC
HES
Counts all L2 prefetches for both
code and data.
24H
AAH
L2_RQSTS.MISS
Counts all L2 misses for both code
and data.
24H
FFH
L2_RQSTS.REFEREN
CES
Counts all L2 requests for both code
and data.
26H
01H
L2_DATA_RQSTS.DE
MAND.I_STATE
Counts number of L2 data demand
loads where the cache line to be
loaded is in the I (invalid) state, i.e. a
cache miss. L2 demand loads are
both L1D demand misses and L1D
prefetches.
Description
Comment
Counts all L2 store RFO requests. L2
RFO requests include both L1D
demand RFO misses as well as L1D
RFO prefetches.
Vol. 3B A-25
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
26H
02H
L2_DATA_RQSTS.DE
MAND.S_STATE
Counts number of L2 data demand
loads where the cache line to be
loaded is in the S (shared) state. L2
demand loads are both L1D demand
misses and L1D prefetches.
26H
04H
L2_DATA_RQSTS.DE
MAND.E_STATE
Counts number of L2 data demand
loads where the cache line to be
loaded is in the E (exclusive) state.
L2 demand loads are both L1D
demand misses and L1D prefetches.
26H
08H
L2_DATA_RQSTS.DE
MAND.M_STATE
Counts number of L2 data demand
loads where the cache line to be
loaded is in the M (modified) state.
L2 demand loads are both L1D
demand misses and L1D prefetches.
26H
0FH
L2_DATA_RQSTS.DE
MAND.MESI
Counts all L2 data demand requests.
L2 demand loads are both L1D
demand misses and L1D prefetches.
26H
10H
L2_DATA_RQSTS.PR
EFETCH.I_STATE
Counts number of L2 prefetch data
loads where the cache line to be
loaded is in the I (invalid) state, i.e. a
cache miss.
26H
20H
L2_DATA_RQSTS.PR
EFETCH.S_STATE
Counts number of L2 prefetch data
loads where the cache line to be
loaded is in the S (shared) state. A
prefetch RFO will miss on an S state
line, while a prefetch read will hit on
an S state line.
26H
40H
L2_DATA_RQSTS.PR
EFETCH.E_STATE
Counts number of L2 prefetch data
loads where the cache line to be
loaded is in the E (exclusive) state.
26H
80H
L2_DATA_RQSTS.PR
EFETCH.M_STATE
Counts number of L2 prefetch data
loads where the cache line to be
loaded is in the M (modified) state.
26H
F0H
L2_DATA_RQSTS.PR
EFETCH.MESI
Counts all L2 prefetch requests.
26H
FFH
L2_DATA_RQSTS.AN Counts all L2 data requests.
Y
A-26 Vol. 3B
Description
Comment
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
27H
01H
L2_WRITE.RFO.I_STA Counts number of L2 demand store This is a demand
TE
RFO requests where the cache line RFO request
to be loaded is in the I (invalid) state,
i.e, a cache miss. The L1D prefetcher
does not issue a RFO prefetch.
27H
02H
L2_WRITE.RFO.S_ST
ATE
27H
08H
L2_WRITE.RFO.M_ST Counts number of L2 store RFO
This is a demand
ATE
requests where the cache line to be RFO request
loaded is in the M (modified) state.
The L1D prefetcher does not issue a
RFO prefetch.
27H
0EH
L2_WRITE.RFO.HIT
This is a demand
Counts number of L2 store RFO
requests where the cache line to be RFO request
loaded is in either the S, E or M
states. The L1D prefetcher does not
issue a RFO prefetch.
27H
0FH
L2_WRITE.RFO.MESI
Counts all L2 store RFO
requests.The L1D prefetcher does
not issue a RFO prefetch.
27H
10H
L2_WRITE.LOCK.I_ST Counts number of L2 demand lock
ATE
RFO requests where the cache line
to be loaded is in the I (invalid) state,
i.e. a cache miss.
27H
20H
L2_WRITE.LOCK.S_S
TATE
Counts number of L2 lock RFO
requests where the cache line to be
loaded is in the S (shared) state.
27H
40H
L2_WRITE.LOCK.E_S
TATE
Counts number of L2 demand lock
RFO requests where the cache line
to be loaded is in the E (exclusive)
state.
27H
80H
L2_WRITE.LOCK.M_S
TATE
Counts number of L2 demand lock
RFO requests where the cache line
to be loaded is in the M (modified)
state.
Description
Comment
Counts number of L2 store RFO
This is a demand
requests where the cache line to be RFO request
loaded is in the S (shared) state. The
L1D prefetcher does not issue a
RFO prefetch,.
This is a demand
RFO request
Vol. 3B A-27
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
27H
E0H
L2_WRITE.LOCK.HIT
27H
F0H
L2_WRITE.LOCK.MESI Counts all L2 demand lock RFO
requests.
28H
01H
L1D_WB_L2.I_STATE Counts number of L1 writebacks to
the L2 where the cache line to be
written is in the I (invalid) state, i.e.
a cache miss.
28H
02H
L1D_WB_L2.S_STAT
E
Counts number of L1 writebacks to
the L2 where the cache line to be
written is in the S state.
28H
04H
L1D_WB_L2.E_STAT
E
Counts number of L1 writebacks to
the L2 where the cache line to be
written is in the E (exclusive) state.
28H
08H
L1D_WB_L2.M_STAT Counts number of L1 writebacks to
E
the L2 where the cache line to be
written is in the M (modified) state.
28H
0FH
L1D_WB_L2.MESI
2EH
4FH
L3_LAT_CACHE.REFE This event counts requests
see Table A-1
RENCE
originating from the core that
reference a cache line in the last
level cache. The event count
includes speculative traffic but
excludes cache line fills due to a L2
hardware-prefetch. Because cache
hierarchy, cache sizes and other
implementation-specific
characteristics; value comparison to
estimate performance differences is
not recommended.
A-28 Vol. 3B
Description
Comment
Counts number of L2 demand lock
RFO requests where the cache line
to be loaded is in either the S, E, or
M state.
Counts all L1 writebacks to the L2 .
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
2EH
41H
L3_LAT_CACHE.MISS This event counts each cache miss see Table A-1
condition for references to the last
level cache. The event count may
include speculative traffic but
excludes cache line fills due to L2
hardware-prefetches. Because
cache hierarchy, cache sizes and
other implementation-specific
characteristics; value comparison to
estimate performance differences is
not recommended.
3CH
00H
CPU_CLK_UNHALTED Counts the number of thread cycles see Table A-1
.THREAD_P
while the thread is not in a halt
state. The thread enters the halt
state when it is running the HLT
instruction. The core frequency may
change from time to time due to
power or thermal throttling.
3CH
01H
CPU_CLK_UNHALTED Increments at the frequency of TSC see Table A-1
.REF_P
when not halted.
40H
01H
L1D_CACHE_LD.I_ST
ATE
40H
02H
L1D_CACHE_LD.S_ST Counts L1 data cache read requests Counter 0, 1 only
ATE
where the cache line to be loaded is
in the S (shared) state.
40H
04H
L1D_CACHE_LD.E_ST Counts L1 data cache read requests Counter 0, 1 only
ATE
where the cache line to be loaded is
in the E (exclusive) state.
40H
08H
L1D_CACHE_LD.M_S
TATE
40H
0FH
L1D_CACHE_LD.MESI Counts L1 data cache read requests. Counter 0, 1 only
41H
02H
L1D_CACHE_ST.S_ST Counts L1 data cache store RFO
Counter 0, 1 only
ATE
requests where the cache line to be
loaded is in the S (shared) state.
Description
Comment
Counts L1 data cache read requests Counter 0, 1 only
where the cache line to be loaded is
in the I (invalid) state, i.e. the read
request missed the cache.
Counts L1 data cache read requests Counter 0, 1 only
where the cache line to be loaded is
in the M (modified) state.
Vol. 3B A-29
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
41H
04H
L1D_CACHE_ST.E_ST Counts L1 data cache store RFO
Counter 0, 1 only
ATE
requests where the cache line to be
loaded is in the E (exclusive) state.
41H
08H
L1D_CACHE_ST.M_S
TATE
Counts L1 data cache store RFO
requests where cache line to be
loaded is in the M (modified) state.
Counter 0, 1 only
42H
01H
L1D_CACHE_LOCK.HI Counts retired load locks that hit in
T
the L1 data cache or hit in an
already allocated fill buffer. The
lock portion of the load lock
transaction must hit in the L1D.
The initial load
will pull the lock
into the L1 data
cache. Counter 0,
1 only
42H
02H
L1D_CACHE_LOCK.S_ Counts L1 data cache retired load
Counter 0, 1 only
STATE
locks that hit the target cache line in
the shared state.
42H
04H
L1D_CACHE_LOCK.E_ Counts L1 data cache retired load
Counter 0, 1 only
STATE
locks that hit the target cache line in
the exclusive state.
42H
08H
L1D_CACHE_LOCK.M
_STATE
Counts L1 data cache retired load
Counter 0, 1 only
locks that hit the target cache line in
the modified state.
43H
01H
L1D_ALL_REF.ANY
Counts all references (uncached,
speculated and retired) to the L1
data cache, including all loads and
stores with any memory types. The
event counts memory accesses only
when they are actually performed.
For example, a load blocked by
unknown store address and later
performed is only counted once.
43H
02H
L1D_ALL_REF.CACHE Counts all data reads and writes
Counter 0, 1 only
ABLE
(speculated and retired) from
cacheable memory, including locked
operations.
49H
01H
DTLB_MISSES.ANY
49H
02H
DTLB_MISSES.WALK_ Counts number of misses in the
COMPLETED
STLB which resulted in a completed
page walk.
A-30 Vol. 3B
Description
Counts the number of misses in the
STLB which causes a page walk.
Comment
The event does
not include nonmemory
accesses, such as
I/O accesses.
Counter 0, 1 only
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
49H
10H
DTLB_MISSES.STLB_ Counts the number of DTLB first
HIT
level misses that hit in the second
level TLB. This event is only
relevant if the core contains
multiple DTLB levels.
49H
20H
DTLB_MISSES.PDE_M Number of DTLB misses caused by
ISS
low part of address, includes
references to 2M pages because 2M
pages do not use the PDE.
49H
80H
DTLB_MISSES.LARGE Counts number of misses in the
_WALK_COMPLETED STLB which resulted in a completed
page walk for large pages.
4CH
01H
LOAD_HIT_PRE
Counts load operations sent to the
L1 data cache while a previous SSE
prefetch instruction to the same
cache line has started prefetching
but has not yet finished.
4EH
01H
L1D_PREFETCH.REQ
UESTS
Counts number of hardware
prefetch requests dispatched out of
the prefetch FIFO.
4EH
02H
L1D_PREFETCH.MISS Counts number of hardware
prefetch requests that miss the
L1D. There are two prefetchers in
the L1D. A streamer, which predicts
lines sequentially after this one
should be fetched, and the IP
prefetcher that remembers access
patterns for the current instruction.
The streamer prefetcher stops on
an L1D hit, while the IP prefetcher
does not.
4EH
04H
L1D_PREFETCH.TRIG Counts number of prefetch requests
GERS
triggered by the Finite State
Machine and pushed into the
prefetch FIFO. Some of the prefetch
requests are dropped due to
overwrites or competition between
the IP index prefetcher and
streamer prefetcher. The prefetch
FIFO contains 4 entries.
Description
Comment
Vol. 3B A-31
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
51H
01H
L1D.REPL
Counts the number of lines brought Counter 0, 1 only
into the L1 data cache.
51H
02H
L1D.M_REPL
Counts the number of modified lines Counter 0, 1 only
brought into the L1 data cache.
51H
04H
L1D.M_EVICT
Counts the number of modified lines Counter 0, 1 only
evicted from the L1 data cache due
to replacement.
51H
08H
L1D.M_SNOOP_EVIC
T
Counts the number of modified lines Counter 0, 1 only
evicted from the L1 data cache due
to snoop HITM intervention.
52H
01H
L1D_CACHE_PREFET Counts the number of cacheable
CH_LOCK_FB_HIT
load lock speculated instructions
accepted into the fill buffer.
53H
01H
L1D_CACHE_LOCK_F Counts the number of cacheable
B_HIT
load lock speculated or retired
instructions accepted into the fill
buffer.
63H
01H
CACHE_LOCK_CYCLE
S.L1D_L2
Cycle count during which the L1D
and L2 are locked. A lock is
asserted when there is a locked
memory access, due to uncacheable
memory, a locked operation that
spans two cache lines, or a page
walk from an uncacheable page
table.
Counter 0, 1 only.
L1D and L2 locks
have a very high
performance
penalty and it is
highly
recommended to
avoid such
accesses.
63H
02H
CACHE_LOCK_CYCLE
S.L1D
Counts the number of cycles that
cacheline in the L1 data cache unit
is locked.
Counter 0, 1 only.
6CH
01H
IO_TRANSACTIONS
Counts the number of completed I/O
transactions.
80H
01H
L1I.HITS
Counts all instruction fetches that
hit the L1 instruction cache.
A-32 Vol. 3B
Description
Comment
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
80H
02H
L1I.MISSES
Counts all instruction fetches that
miss the L1I cache. This includes
instruction cache misses, streaming
buffer misses, victim cache misses
and uncacheable fetches. An
instruction fetch miss is counted
only once and not once for every
cycle it is outstanding.
80H
03H
L1I.READS
Counts all instruction fetches,
including uncacheable fetches that
bypass the L1I.
80H
04H
L1I.CYCLES_STALLED Cycle counts for which an
instruction fetch stalls due to a L1I
cache miss, ITLB miss or ITLB fault.
82H
01H
LARGE_ITLB.HIT
Counts number of large ITLB hits.
85H
01H
ITLB_MISSES.ANY
Counts the number of misses in all
levels of the ITLB which causes a
page walk.
85H
02H
ITLB_MISSES.WALK_
COMPLETED
Counts number of misses in all
levels of the ITLB which resulted in
a completed page walk.
87H
01H
ILD_STALL.LCP
Cycles Instruction Length Decoder
stalls due to length changing
prefixes: 66, 67 or REX.W (for
EM64T) instructions which change
the length of the decoded
instruction.
87H
02H
ILD_STALL.MRU
Instruction Length Decoder stall
cycles due to Brand Prediction Unit
(PBU) Most Recently Used (MRU)
bypass.
87H
04H
ILD_STALL.IQ_FULL
Stall cycles due to a full instruction
queue.
87H
08H
ILD_STALL.REGEN
Counts the number of regen stalls.
87H
0FH
ILD_STALL.ANY
Counts any cycles the Instruction
Length Decoder is stalled.
Description
Comment
Vol. 3B A-33
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
88H
01H
BR_INST_EXEC.COND Counts the number of conditional
near branch instructions executed,
but not necessarily retired.
88H
02H
BR_INST_EXEC.DIRE
CT
88H
04H
BR_INST_EXEC.INDIR Counts the number of executed
ECT_NON_CALL
indirect near branch instructions
that are not calls.
88H
07H
BR_INST_EXEC.NON
_CALLS
88H
08H
BR_INST_EXEC.RETU Counts indirect near branches that
RN_NEAR
have a return mnemonic.
88H
10H
BR_INST_EXEC.DIRE
CT_NEAR_CALL
88H
20H
BR_INST_EXEC.INDIR Counts indirect near calls, including
ECT_NEAR_CALL
both register and memory indirect,
executed.
88H
30H
BR_INST_EXEC.NEAR Counts all near call branches
_CALLS
executed, but not necessarily
retired.
88H
40H
BR_INST_EXEC.TAKE Counts taken near branches
N
executed, but not necessarily
retired.
88H
7FH
BR_INST_EXEC.ANY
Counts all near executed branches
(not necessarily retired). This
includes only instructions and not
micro-op branches. Frequent
branching is not necessarily a major
performance issue. However
frequent branch mispredictions may
be a problem.
89H
01H
BR_MISP_EXEC.CON
D
Counts the number of mispredicted
conditional near branch instructions
executed, but not necessarily
retired.
A-34 Vol. 3B
Description
Counts all unconditional near branch
instructions excluding calls and
indirect branches.
Counts all non call near branch
instructions executed, but not
necessarily retired.
Counts unconditional near call
branch instructions, excluding non
call branch, executed.
Comment
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
89H
02H
BR_MISP_EXEC.DIRE
CT
89H
04H
BR_MISP_EXEC.INDIR Counts the number of executed
ECT_NON_CALL
mispredicted indirect near branch
instructions that are not calls.
89H
07H
BR_MISP_EXEC.NON
_CALLS
89H
08H
BR_MISP_EXEC.RETU Counts mispredicted indirect
RN_NEAR
branches that have a rear return
mnemonic.
89H
10H
BR_MISP_EXEC.DIRE
CT_NEAR_CALL
89H
20H
BR_MISP_EXEC.INDIR Counts mispredicted indirect near
ECT_NEAR_CALL
calls exeucted, including both
register and memory indirect.
89H
30H
BR_MISP_EXEC.NEA
R_CALLS
89H
40H
BR_MISP_EXEC.TAKE Counts executed mispredicted near
N
branches that are taken, but not
necessarily retired.
89H
7FH
BR_MISP_EXEC.ANY
Description
Comment
Counts mispredicted macro
unconditional near branch
instructions, excluding calls and
indirect branches (should always be
0).
Counts mispredicted non call near
branches executed, but not
necessarily retired.
Counts mispredicted non-indirect
near calls executed, (should always
be 0).
Counts all mispredicted near call
branches executed, but not
necessarily retired.
Counts the number of mispredicted
near branch instructions that were
executed, but not necessarily
retired.
Vol. 3B A-35
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
A2H
01H
RESOURCE_STALLS.
ANY
A2H
02H
RESOURCE_STALLS.L Counts the cycles of stall due to lack
OAD
of load buffer for load operation.
A2H
04H
RESOURCE_STALLS.R This event counts the number of
S_FULL
cycles when the number of
instructions in the pipeline waiting
for execution reaches the limit the
processor can handle. A high count
of this event indicates that there
are long latency operations in the
pipe (possibly load and store
operations that miss the L2 cache,
or instructions dependent upon
instructions further down the
pipeline that have yet to retire.
A2H
08H
RESOURCE_STALLS.S This event counts the number of
TORE
cycles that a resource related stall
will occur due to the number of
store instructions reaching the limit
of the pipeline, (i.e. all store buffers
are used). The stall ends when a
store instruction commits its data to
the cache or memory.
A2H
10H
RESOURCE_STALLS.R Counts the cycles of stall due to reOB_FULL
order buffer full.
A2H
20H
RESOURCE_STALLS.F Counts the number of cycles while
PCW
execution was stalled due to writing
the floating-point unit (FPU) control
word.
A-36 Vol. 3B
Description
Comment
Counts the number of Allocator
resource related stalls. Includes
register renaming buffer entries,
memory buffer entries. In addition
to resource related stalls, this event
counts some other events. Includes
stalls arising during branch
misprediction recovery, such as if
retirement of the mispredicted
branch is delayed and stalls arising
while store buffer is draining from
synchronizing operations.
Does not include
stalls due to
SuperQ (off core)
queue full, too
many cache
misses, etc.
When RS is full,
new instructions
can not enter the
reservation
station and start
execution.
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
A2H
40H
RESOURCE_STALLS.
MXCSR
Stalls due to the MXCSR register
rename occurring to close to a
previous MXCSR rename. The
MXCSR provides control and status
for the MMX registers.
A2H
80H
RESOURCE_STALLS.
OTHER
Counts the number of cycles while
execution was stalled due to other
resource issues.
A6H
01H
MACRO_INSTS.FUSIO Counts the number of instructions
NS_DECODED
decoded that are macro-fused but
not necessarily executed or retired.
A7H
01H
BACLEAR_FORCE_IQ
Counts number of times a BACLEAR
was forced by the Instruction
Queue. The IQ is also responsible
for providing conditional branch
prediciton direction based on a
static scheme and dynamic data
provided by the L2 Branch
Prediction Unit. If the conditional
branch target is not found in the
Target Array and the IQ predicts
that the branch is taken, then the IQ
will force the Branch Address
Calculator to issue a BACLEAR. Each
BACLEAR asserted by the BAC
generates approximately an 8 cycle
bubble in the instruction fetch
pipeline.
A8H
01H
LSD.UOPS
Counts the number of micro-ops
delivered by loop stream detector.
AEH
01H
ITLB_FLUSH
Counts the number of ITLB flushes.
B0H
40H
OFFCORE_REQUEST
S.L1D_WRITEBACK
Counts number of L1D writebacks
to the uncore.
B1H
01H
UOPS_EXECUTED.PO Counts number of Uops executed
RT0
that were issued on port 0. Port 0
handles integer arithmetic, SIMD
and FP add Uops.
Description
Comment
Use cmask=1 and
invert to count
cycles
Vol. 3B A-37
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
B1H
02H
UOPS_EXECUTED.PO Counts number of Uops executed
RT1
that were issued on port 1. Port 1
handles integer arithmetic, SIMD,
integer shift, FP multiply and FP
divide Uops.
B1H
04H
UOPS_EXECUTED.PO Counts number of Uops executed
RT2_CORE
that were issued on port 2. Port 2
handles the load Uops. This is a core
count only and can not be collected
per thread.
B1H
08H
UOPS_EXECUTED.PO Counts number of Uops executed
RT3_CORE
that were issued on port 3. Port 3
handles store Uops. This is a core
count only and can not be collected
per thread.
B1H
10H
UOPS_EXECUTED.PO Counts number of Uops executed
RT4_CORE
that where issued on port 4. Port 4
handles the value to be stored for
the store Uops issued on port 3.
This is a core count only and can not
be collected per thread.
B1H
1FH
UOPS_EXECUTED.CO Counts cycles when the Uops
RE_ACTIVE_CYCLES_ executed were issued from any
NO_PORT5
ports except port 5. Use Cmask=1
for active cycles; Cmask=0 for
weighted cycles; Use CMask=1,
Invert=1 to count P0-4 stalled
cycles Use Cmask=1, Edge=1,
Invert=1 to count P0-4 stalls.
B1H
20H
UOPS_EXECUTED.PO Counts number of Uops executed
RT5
that where issued on port 5.
B1H
3FH
UOPS_EXECUTED.CO Counts cycles when the Uops are
RE_ACTIVE_CYCLES executing . Use Cmask=1 for active
cycles; Cmask=0 for weighted
cycles; Use CMask=1, Invert=1 to
count P0-4 stalled cycles Use
Cmask=1, Edge=1, Invert=1 to
count P0-4 stalls.
A-38 Vol. 3B
Description
Comment
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
B1H
40H
UOPS_EXECUTED.PO Counts number of Uops executed
use cmask=1,
RT015
that where issued on port 0, 1, or 5. invert=1 to count
stall cycles
B1H
80H
UOPS_EXECUTED.PO Counts number of Uops executed
RT234
that where issued on port 2, 3, or 4.
B2H
01H
OFFCORE_REQUEST
S_SQ_FULL
B7H
01H
OFF_CORE_RESPONS see Section 30.6.1.3, “Off-core
E_0
Response Performance Monitoring
in the Processor Core”.
B8H
01H
SNOOP_RESPONSE.H Counts HIT snoop response sent by
IT
this thread in response to a snoop
request.
B8H
02H
SNOOP_RESPONSE.H Counts HIT E snoop response sent
ITE
by this thread in response to a
snoop request.
B8H
04H
SNOOP_RESPONSE.H Counts HIT M snoop response sent
ITM
by this thread in response to a
snoop request.
BBH
01H
Requires
OFF_CORE_RESPONS See Section 30.7, “Performance
E_1
Monitoring for Processors Based on programming
MSR 01A7H
Intel® Microarchitecture Code
Name Westmere”.
C0H
01H
INST_RETIRED.ANY_ See Table A-1
P
Notes: INST_RETIRED.ANY is
counted by a designated fixed
counter. INST_RETIRED.ANY_P is
counted by a programmable counter
and is an architectural performance
event. Event is supported if
CPUID.A.EBX[1] = 0.
C0H
02H
INST_RETIRED.X87
Description
Comment
Counts number of cycles the SQ is
full to handle off-core requests.
Requires
programming
MSR 01A6H
Counting:
Faulting
executions of
GETSEC/VM
entry/VM
Exit/MWait will
not count as
retired
instructions.
Counts the number of MMX
instructions retired.
Vol. 3B A-39
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
C0H
04H
INST_RETIRED.MMX
Counts the number of floating point
computational operations retired:
floating point computational
operations executed by the assist
handler and sub-operations of
complex floating point instructions
like transcendental instructions.
C2H
01H
UOPS_RETIRED.ANY
Counts the number of micro-ops
retired, (macro-fused=1, microfused=2, others=1; maximum count
of 8 per cycle). Most instructions are
composed of one or two micro-ops.
Some instructions are decoded into
longer sequences such as repeat
instructions, floating point
transcendental instructions, and
assists.
C2H
02H
UOPS_RETIRED.RETI
RE_SLOTS
Counts the number of retirement
slots used each cycle.
C2H
04H
UOPS_RETIRED.MAC
RO_FUSED
Counts number of macro-fused uops
retired.
C3H
01H
MACHINE_CLEARS.CY Counts the cycles machine clear is
CLES
asserted.
C3H
02H
MACHINE_CLEARS.M
EM_ORDER
Counts the number of machine
clears due to memory order
conflicts.
C3H
04H
MACHINE_CLEARS.S
MC
Counts the number of times that a
program writes to a code section.
Self-modifying code causes a sever
penalty in all Intel 64 and IA-32
processors. The modified cache line
is written back to the L2 and
L3caches.
C4H
00H
BR_INST_RETIRED.A
LL_BRANCHES
Branch instructions at retirement
C4H
01H
BR_INST_RETIRED.C
ONDITIONAL
Counts the number of conditional
branch instructions retired.
A-40 Vol. 3B
Description
Comment
Use cmask=1 and
invert to count
active cycles or
stalled cycles
See Table A-1
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
C4H
02H
BR_INST_RETIRED.N Counts the number of direct &
EAR_CALL
indirect near unconditional calls
retired.
C4H
04H
BR_INST_RETIRED.A
LL_BRANCHES
C5H
00H
BR_MISP_RETIRED.A Mispredicted branch instructions at
LL_BRANCHES
retirement
C5H
02H
BR_MISP_RETIRED.N Counts mispredicted direct &
EAR_CALL
indirect near unconditional retired
calls.
C7H
01H
SSEX_UOPS_RETIRE
D.PACKED_SINGLE
Counts SIMD packed single-precision
floating point Uops retired.
C7H
02H
SSEX_UOPS_RETIRE
D.SCALAR_SINGLE
Counts SIMD calar single-precision
floating point Uops retired.
C7H
04H
SSEX_UOPS_RETIRE
D.PACKED_DOUBLE
Counts SIMD packed doubleprecision floating point Uops retired.
C7H
08H
SSEX_UOPS_RETIRE
D.SCALAR_DOUBLE
Counts SIMD scalar double-precision
floating point Uops retired.
C7H
10H
SSEX_UOPS_RETIRE
D.VECTOR_INTEGER
Counts 128-bit SIMD vector integer
Uops retired.
C8H
20H
ITLB_MISS_RETIRED
Counts the number of retired
instructions that missed the ITLB
when the instruction was fetched.
CBH
01H
MEM_LOAD_RETIRED Counts number of retired loads that
.L1D_HIT
hit the L1 data cache.
CBH
02H
MEM_LOAD_RETIRED Counts number of retired loads that
.L2_HIT
hit the L2 data cache.
CBH
04H
MEM_LOAD_RETIRED Counts number of retired loads that
.L3_UNSHARED_HIT hit their own, unshared lines in the
L3 cache.
CBH
08H
MEM_LOAD_RETIRED Counts number of retired loads that
.OTHER_CORE_L2_HI hit in a sibling core's L2 (on die core).
T_HITM
Since the L3 is inclusive of all cores
on the package, this is an L3 hit.
This counts both clean or modified
hits.
Description
Comment
Counts the number of branch
instructions retired.
See Table A-1
Vol. 3B A-41
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
CBH
10H
MEM_LOAD_RETIRED Counts number of retired loads that
.L3_MISS
miss the L3 cache. The load was
satisfied by a remote socket, local
memory or an IOH.
CBH
40H
MEM_LOAD_RETIRED Counts number of retired loads that
.HIT_LFB
miss the L1D and the address is
located in an allocated line fill buffer
and will soon be committed to
cache. This is counting secondary
L1D misses.
CBH
80H
MEM_LOAD_RETIRED Counts the number of retired loads
.DTLB_MISS
that missed the DTLB. The DTLB
miss is not counted if the load
operation causes a fault. This event
counts loads from cacheable
memory only. The event does not
count loads by software prefetches.
Counts both primary and secondary
misses to the TLB.
CCH
01H
FP_MMX_TRANS.TO
_FP
Counts the first floating-point
instruction following any MMX
instruction. You can use this event
to estimate the penalties for the
transitions between floating-point
and MMX technology states.
CCH
02H
FP_MMX_TRANS.TO
_MMX
Counts the first MMX instruction
following a floating-point
instruction. You can use this event
to estimate the penalties for the
transitions between floating-point
and MMX technology states.
CCH
03H
FP_MMX_TRANS.AN
Y
Counts all transitions from floating
point to MMX instructions and from
MMX instructions to floating point
instructions. You can use this event
to estimate the penalties for the
transitions between floating-point
and MMX technology states.
A-42 Vol. 3B
Description
Comment
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
D0H
01H
MACRO_INSTS.DECO
DED
Counts the number of instructions
decoded, (but not necessarily
executed or retired).
D1H
02H
UOPS_DECODED.MS
Counts the number of Uops decoded
by the Microcode Sequencer, MS.
The MS delivers uops when the
instruction is more than 4 uops long
or a microcode assist is occurring.
D1H
04H
UOPS_DECODED.ESP Counts number of stack pointer
_FOLDING
(ESP) instructions decoded: push ,
pop , call , ret, etc. ESP instructions
do not generate a Uop to increment
or decrement ESP. Instead, they
update an ESP_Offset register that
keeps track of the delta to the
current value of the ESP register.
D1H
08H
UOPS_DECODED.ESP Counts number of stack pointer
_SYNC
(ESP) sync operations where an ESP
instruction is corrected by adding
the ESP offset register to the
current value of the ESP register.
D2H
01H
RAT_STALLS.FLAGS
Counts the number of cycles during
which execution stalled due to
several reasons, one of which is a
partial flag register stall. A partial
register stall may occur when two
conditions are met: 1) an instruction
modifies some, but not all, of the
flags in the flag register and 2) the
next instruction, which depends on
flags, depends on flags that were
not modified by this instruction.
D2H
02H
RAT_STALLS.REGIST
ERS
This event counts the number of
cycles instruction execution latency
became longer than the defined
latency because the instruction
used a register that was partially
written by previous instruction.
Description
Comment
Vol. 3B A-43
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
D2H
04H
RAT_STALLS.ROB_RE Counts the number of cycles when
AD_PORT
ROB read port stalls occurred, which
did not allow new micro-ops to
enter the out-of-order pipeline.
Note that, at this stage in the
pipeline, additional stalls may occur
at the same cycle and prevent the
stalled micro-ops from entering the
pipe. In such a case, micro-ops retry
entering the execution pipe in the
next cycle and the ROB-read port
stall is counted again.
D2H
08H
RAT_STALLS.SCOREB Counts the cycles where we stall
OARD
due to microarchitecturally required
serialization. Microcode
scoreboarding stalls.
D2H
0FH
RAT_STALLS.ANY
D4H
01H
SEG_RENAME_STALL Counts the number of stall cycles
S
due to the lack of renaming
resources for the ES, DS, FS, and GS
segment registers. If a segment is
renamed but not retired and a
second update to the same
segment occurs, a stall occurs in the
front-end of the pipeline until the
renamed segment retires.
A-44 Vol. 3B
Description
Counts all Register Allocation Table
stall cycles due to: Cycles when
ROB read port stalls occurred, which
did not allow new micro-ops to
enter the execution pipe. Cycles
when partial register stalls occurred
Cycles when flag stalls occurred
Cycles floating-point unit (FPU)
status word stalls occurred. To
count each of these conditions
separately use the events:
RAT_STALLS.ROB_READ_PORT,
RAT_STALLS.PARTIAL,
RAT_STALLS.FLAGS, and
RAT_STALLS.FPSW.
Comment
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
D5H
01H
ES_REG_RENAMES
Counts the number of times the ES
segment register is renamed.
DBH
01H
UOP_UNFUSION
Counts unfusion events due to
floating point exception to a fused
uop.
E0H
01H
BR_INST_DECODED
Counts the number of branch
instructions decoded.
E5H
01H
BPU_MISSED_CALL_
RET
Counts number of times the Branch
Prediciton Unit missed predicting a
call or return branch.
E6H
01H
BACLEAR.CLEAR
Counts the number of times the
front end is resteered, mainly when
the Branch Prediction Unit cannot
provide a correct prediction and this
is corrected by the Branch Address
Calculator at the front end. This can
occur if the code has many branches
such that they cannot be consumed
by the BPU. Each BACLEAR asserted
by the BAC generates
approximately an 8 cycle bubble in
the instruction fetch pipeline. The
effect on total execution time
depends on the surrounding code.
E6H
02H
BACLEAR.BAD_TARG Counts number of Branch Address
ET
Calculator clears (BACLEAR)
asserted due to conditional branch
instructions in which there was a
target hit but the direction was
wrong. Each BACLEAR asserted by
the BAC generates approximately
an 8 cycle bubble in the instruction
fetch pipeline.
E8H
01H
BPU_CLEARS.EARLY
Description
Counts early (normal) Branch
Prediction Unit clears: BPU
predicted a taken branch after
incorrectly assuming that it was not
taken.
Comment
The BPU clear
leads to 2 cycle
bubble in the
Front End.
Vol. 3B A-45
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
E8H
02H
BPU_CLEARS.LATE
F0H
01H
L2_TRANSACTIONS.L Counts L2 load operations due to
OAD
HW prefetch or demand loads.
F0H
02H
L2_TRANSACTIONS.
RFO
F0H
04H
L2_TRANSACTIONS.I Counts L2 instruction fetch
FETCH
operations due to HW prefetch or
demand ifetch.
F0H
08H
L2_TRANSACTIONS.
PREFETCH
F0H
10H
L2_TRANSACTIONS.L Counts L1D writeback operations to
1D_WB
the L2.
F0H
20H
L2_TRANSACTIONS.
FILL
Counts L2 cache line fill operations
due to load, RFO, L1D writeback or
prefetch.
F0H
40H
L2_TRANSACTIONS.
WB
Counts L2 writeback operations to
the L3.
F0H
80H
L2_TRANSACTIONS.
ANY
Counts all L2 cache operations.
F1H
02H
L2_LINES_IN.S_STAT Counts the number of cache lines
E
allocated in the L2 cache in the S
(shared) state.
F1H
04H
L2_LINES_IN.E_STAT Counts the number of cache lines
E
allocated in the L2 cache in the E
(exclusive) state.
F1H
07H
L2_LINES_IN.ANY
F2H
01H
L2_LINES_OUT.DEMA Counts L2 clean cache lines evicted
ND_CLEAN
by a demand request.
F2H
02H
L2_LINES_OUT.DEMA Counts L2 dirty (modified) cache
ND_DIRTY
lines evicted by a demand request.
F2H
04H
L2_LINES_OUT.PREF Counts L2 clean cache line evicted
ETCH_CLEAN
by a prefetch request.
A-46 Vol. 3B
Description
Counts late Branch Prediction Unit
clears due to Most Recently Used
conflicts. The PBU clear leads to a 3
cycle bubble in the Front End.
Counts L2 RFO operations due to
HW prefetch or demand RFOs.
Counts L2 prefetch operations.
Counts the number of cache lines
allocated in the L2 cache.
Comment
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
F2H
08H
L2_LINES_OUT.PREF Counts L2 modified cache line
ETCH_DIRTY
evicted by a prefetch request.
F2H
0FH
L2_LINES_OUT.ANY
F4H
10H
SQ_MISC.SPLIT_LOCK Counts the number of SQ lock splits
across a cache line.
F6H
01H
SQ_FULL_STALL_CY
CLES
Counts cycles the Super Queue is
full. Neither of the threads on this
core will be able to access the
uncore.
F7H
01H
FP_ASSIST.ALL
Counts the number of floating point
operations executed that required
micro-code assist intervention.
Assists are required in the following
cases: SSE instructions, (Denormal
input when the DAZ flag is off or
Underflow result when the FTZ flag
is off): x87 instructions, (NaN or
denormal are loaded to a register or
used as input from memory, Division
by 0 or Underflow output).
F7H
02H
FP_ASSIST.OUTPUT
Counts number of floating point
micro-code assist when the output
value (destination register) is
invalid.
F7H
04H
FP_ASSIST.INPUT
Counts number of floating point
micro-code assist when the input
value (one of the source operands
to an FP instruction) is invalid.
FDH
01H
SIMD_INT_64.PACKE
D_MPY
Counts number of SID integer 64 bit
packed multiply operations.
FDH
02H
SIMD_INT_64.PACKE
D_SHIFT
Counts number of SID integer 64 bit
packed shift operations.
FDH
04H
SIMD_INT_64.PACK
Counts number of SID integer 64 bit
pack operations.
FDH
08H
SIMD_INT_64.UNPAC Counts number of SID integer 64 bit
K
unpack operations.
Description
Comment
Counts all L2 cache lines evicted for
any reason.
Vol. 3B A-47
PERFORMANCE-MONITORING EVENTS
Table A-4. Non-Architectural Performance Events In the Processor Core for Intel Core
i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
Description
FDH
10H
SIMD_INT_64.PACKE
D_LOGICAL
Counts number of SID integer 64 bit
logical operations.
FDH
20H
SIMD_INT_64.PACKE
D_ARITH
Counts number of SID integer 64 bit
arithmetic operations.
FDH
40H
SIMD_INT_64.SHUFF Counts number of SID integer 64 bit
LE_MOVE
shift or move operations.
Comment
Non-architectural Performance monitoring events that are located in the uncore subsystem are implementation specific between different platforms using processors
based on Intel microarchitecture code name Nehalem. Processors with CPUID signature of DisplayFamily_DisplayModel 06_1AH, 06_1EH, and 06_1FH support performance events listed in Table A-5.
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
00H
01H
UNC_GQ_CYCLES_FU Uncore cycles Global Queue read
LL.READ_TRACKER
tracker is full.
00H
02H
UNC_GQ_CYCLES_FU Uncore cycles Global Queue write
LL.WRITE_TRACKER tracker is full.
00H
04H
UNC_GQ_CYCLES_FU Uncore cycles Global Queue peer
LL.PEER_PROBE_TR probe tracker is full. The peer probe
ACKER
tracker queue tracks snoops from the
IOH and remote sockets.
01H
01H
UNC_GQ_CYCLES_NO Uncore cycles were Global Queue read
T_EMPTY.READ_TRA tracker has at least one valid entry.
CKER
01H
02H
UNC_GQ_CYCLES_NO Uncore cycles were Global Queue
T_EMPTY.WRITE_TR write tracker has at least one valid
entry.
ACKER
01H
04H
UNC_GQ_CYCLES_NO Uncore cycles were Global Queue peer
T_EMPTY.PEER_PRO probe tracker has at least one valid
BE_TRACKER
entry. The peer probe tracker queue
tracks IOH and remote socket snoops.
A-48 Vol. 3B
Description
Comment
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
03H
01H
UNC_GQ_ALLOC.REA Counts the number of tread tracker
D_TRACKER
allocate to deallocate entries. The GQ
read tracker allocate to deallocate
occupancy count is divided by the
count to obtain the average read
tracker latency.
03H
02H
UNC_GQ_ALLOC.RT_
L3_MISS
Counts the number GQ read tracker
entries for which a full cache line read
has missed the L3. The GQ read
tracker L3 miss to fill occupancy count
is divided by this count to obtain the
average cache line read L3 miss
latency. The latency represents the
time after which the L3 has
determined that the cache line has
missed. The time between a GQ read
tracker allocation and the L3
determining that the cache line has
missed is the average L3 hit latency.
The total L3 cache line read miss
latency is the hit latency + L3 miss
latency.
03H
04H
UNC_GQ_ALLOC.RT_
TO_L3_RESP
Counts the number of GQ read tracker
entries that are allocated in the read
tracker queue that hit or miss the L3.
The GQ read tracker L3 hit occupancy
count is divided by this count to
obtain the average L3 hit latency.
03H
08H
UNC_GQ_ALLOC.RT_
TO_RTID_ACQUIRED
Counts the number of GQ read tracker
entries that are allocated in the read
tracker, have missed in the L3 and
have not acquired a Request
Transaction ID. The GQ read tracker
L3 miss to RTID acquired occupancy
count is divided by this count to
obtain the average latency for a read
L3 miss to acquire an RTID.
Description
Comment
Vol. 3B A-49
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
03H
10H
UNC_GQ_ALLOC.WT_ Counts the number of GQ write
TO_RTID_ACQUIRED tracker entries that are allocated in
the write tracker, have missed in the
L3 and have not acquired a Request
Transaction ID. The GQ write tracker
L3 miss to RTID occupancy count is
divided by this count to obtain the
average latency for a write L3 miss to
acquire an RTID.
03H
20H
UNC_GQ_ALLOC.WRI
TE_TRACKER
Counts the number of GQ write
tracker entries that are allocated in
the write tracker queue that miss the
L3. The GQ write tracker occupancy
count is divided by the this count to
obtain the average L3 write miss
latency.
03H
40H
UNC_GQ_ALLOC.PEE
R_PROBE_TRACKER
Counts the number of GQ peer probe
tracker (snoop) entries that are
allocated in the peer probe tracker
queue that miss the L3. The GQ peer
probe occupancy count is divided by
this count to obtain the average L3
peer probe miss latency.
04H
01H
UNC_GQ_DATA.FROM Cycles Global Queue Quickpath
_QPI
Interface input data port is busy
importing data from the Quickpath
Interface. Each cycle the input port
can transfer 8 or 16 bytes of data.
04H
02H
UNC_GQ_DATA.FROM Cycles Global Queue Quickpath
_QMC
Memory Interface input data port is
busy importing data from the
Quickpath Memory Interface. Each
cycle the input port can transfer 8 or
16 bytes of data.
04H
04H
UNC_GQ_DATA.FROM Cycles GQ L3 input data port is busy
_L3
importing data from the Last Level
Cache. Each cycle the input port can
transfer 32 bytes of data.
A-50 Vol. 3B
Description
Comment
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
04H
08H
UNC_GQ_DATA.FROM Cycles GQ Core 0 and 2 input data
_CORES_02
port is busy importing data from
processor cores 0 and 2. Each cycle
the input port can transfer 32 bytes
of data.
04H
10H
UNC_GQ_DATA.FROM Cycles GQ Core 1 and 3 input data
_CORES_13
port is busy importing data from
processor cores 1 and 3. Each cycle
the input port can transfer 32 bytes
of data.
05H
01H
UNC_GQ_DATA.TO_Q Cycles GQ QPI and QMC output data
PI_QMC
port is busy sending data to the
Quickpath Interface or Quickpath
Memory Interface. Each cycle the
output port can transfer 32 bytes of
data.
05H
02H
UNC_GQ_DATA.TO_L
3
05H
04H
UNC_GQ_DATA.TO_C Cycles GQ Core output data port is
ORES
busy sending data to the Cores. Each
cycle the output port can transfer 32
bytes of data.
06H
01H
UNC_SNP_RESP_TO_ Number of snoop responses to the
LOCAL_HOME.I_STAT local home that L3 does not have the
E
referenced cache line.
06H
02H
UNC_SNP_RESP_TO_ Number of snoop responses to the
LOCAL_HOME.S_STA local home that L3 has the referenced
TE
line cached in the S state.
06H
04H
UNC_SNP_RESP_TO_ Number of responses to code or data
LOCAL_HOME.FWD_S read snoops to the local home that
_STATE
the L3 has the referenced cache line
in the E state. The L3 cache line state
is changed to the S state and the line
is forwarded to the local home in the
S state.
Description
Comment
Cycles GQ L3 output data port is busy
sending data to the Last Level Cache.
Each cycle the output port can
transfer 32 bytes of data.
Vol. 3B A-51
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
06H
08H
UNC_SNP_RESP_TO_ Number of responses to read
LOCAL_HOME.FWD_I invalidate snoops to the local home
_STATE
that the L3 has the referenced cache
line in the M state. The L3 cache line
state is invalidated and the line is
forwarded to the local home in the M
state.
06H
10H
UNC_SNP_RESP_TO_ Number of conflict snoop responses
LOCAL_HOME.CONFLI sent to the local home.
CT
06H
20H
UNC_SNP_RESP_TO_ Number of responses to code or data
LOCAL_HOME.WB
read snoops to the local home that
the L3 has the referenced line cached
in the M state.
07H
01H
UNC_SNP_RESP_TO_ Number of snoop responses to a
REMOTE_HOME.I_ST remote home that L3 does not have
ATE
the referenced cache line.
07H
02H
UNC_SNP_RESP_TO_ Number of snoop responses to a
REMOTE_HOME.S_ST remote home that L3 has the
ATE
referenced line cached in the S state.
07H
04H
UNC_SNP_RESP_TO_ Number of responses to code or data
REMOTE_HOME.FWD read snoops to a remote home that
_S_STATE
the L3 has the referenced cache line
in the E state. The L3 cache line state
is changed to the S state and the line
is forwarded to the remote home in
the S state.
07H
08H
UNC_SNP_RESP_TO_ Number of responses to read
REMOTE_HOME.FWD invalidate snoops to a remote home
_I_STATE
that the L3 has the referenced cache
line in the M state. The L3 cache line
state is invalidated and the line is
forwarded to the remote home in the
M state.
07H
10H
UNC_SNP_RESP_TO_ Number of conflict snoop responses
REMOTE_HOME.CON sent to the local home.
FLICT
A-52 Vol. 3B
Description
Comment
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
07H
20H
UNC_SNP_RESP_TO_ Number of responses to code or data
REMOTE_HOME.WB
read snoops to a remote home that
the L3 has the referenced line cached
in the M state.
07H
24H
UNC_SNP_RESP_TO_ Number of HITM snoop responses to a
REMOTE_HOME.HITM remote home
08H
01H
UNC_L3_HITS.READ
08H
02H
UNC_L3_HITS.WRITE Number of writeback requests that
hit in the L3. Writebacks from the
cores will always result in L3 hits due
to the inclusive property of the L3.
08H
04H
UNC_L3_HITS.PROBE Number of snoops from IOH or remote
sockets that hit in the L3.
08H
03H
UNC_L3_HITS.ANY
Number of reads and writes that hit
the L3.
09H
01H
UNC_L3_MISS.READ
Number of code read, data read and
RFO requests that miss the L3.
09H
02H
UNC_L3_MISS.WRITE Number of writeback requests that
miss the L3. Should always be zero as
writebacks from the cores will always
result in L3 hits due to the inclusive
property of the L3.
09H
04H
UNC_L3_MISS.PROBE Number of snoops from IOH or remote
sockets that miss the L3.
09H
03H
UNC_L3_MISS.ANY
Number of reads and writes that miss
the L3.
0AH
01H
UNC_L3_LINES_IN.M
_STATE
Counts the number of L3 lines
allocated in M state. The only time a
cache line is allocated in the M state is
when the line was forwarded in M
state is forwarded due to a Snoop
Read Invalidate Own request.
0AH
02H
UNC_L3_LINES_IN.E_ Counts the number of L3 lines
STATE
allocated in E state.
0AH
04H
UNC_L3_LINES_IN.S_ Counts the number of L3 lines
STATE
allocated in S state.
Description
Comment
Number of code read, data read and
RFO requests that hit in the L3
Vol. 3B A-53
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
0AH
08H
UNC_L3_LINES_IN.F_ Counts the number of L3 lines
STATE
allocated in F state.
0AH
0FH
UNC_L3_LINES_IN.A
NY
0BH
01H
UNC_L3_LINES_OUT. Counts the number of L3 lines
M_STATE
victimized that were in the M state.
When the victim cache line is in M
state, the line is written to its home
cache agent which can be either local
or remote.
0BH
02H
UNC_L3_LINES_OUT. Counts the number of L3 lines
E_STATE
victimized that were in the E state.
0BH
04H
UNC_L3_LINES_OUT. Counts the number of L3 lines
S_STATE
victimized that were in the S state.
0BH
08H
UNC_L3_LINES_OUT. Counts the number of L3 lines
I_STATE
victimized that were in the I state.
0BH
10H
UNC_L3_LINES_OUT. Counts the number of L3 lines
F_STATE
victimized that were in the F state.
0BH
1FH
UNC_L3_LINES_OUT. Counts the number of L3 lines
ANY
victimized in any state.
20H
01H
UNC_QHL_REQUEST
S.IOH_READS
Counts number of Quickpath Home
Logic read requests from the IOH.
20H
02H
UNC_QHL_REQUEST
S.IOH_WRITES
Counts number of Quickpath Home
Logic write requests from the IOH.
20H
04H
UNC_QHL_REQUEST
S.REMOTE_READS
Counts number of Quickpath Home
Logic read requests from a remote
socket.
20H
08H
UNC_QHL_REQUEST
S.REMOTE_WRITES
Counts number of Quickpath Home
Logic write requests from a remote
socket.
20H
10H
UNC_QHL_REQUEST
S.LOCAL_READS
Counts number of Quickpath Home
Logic read requests from the local
socket.
20H
20H
UNC_QHL_REQUEST
S.LOCAL_WRITES
Counts number of Quickpath Home
Logic write requests from the local
socket.
A-54 Vol. 3B
Description
Counts the number of L3 lines
allocated in any state.
Comment
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
21H
01H
UNC_QHL_CYCLES_F Counts uclk cycles all entries in the
ULL.IOH
Quickpath Home Logic IOH are full.
21H
02H
UNC_QHL_CYCLES_F Counts uclk cycles all entries in the
ULL.REMOTE
Quickpath Home Logic remote tracker
are full.
21H
04H
UNC_QHL_CYCLES_F Counts uclk cycles all entries in the
ULL.LOCAL
Quickpath Home Logic local tracker
are full.
22H
01H
UNC_QHL_CYCLES_N Counts uclk cycles all entries in the
OT_EMPTY.IOH
Quickpath Home Logic IOH is busy.
22H
02H
UNC_QHL_CYCLES_N Counts uclk cycles all entries in the
OT_EMPTY.REMOTE Quickpath Home Logic remote tracker
is busy.
22H
04H
UNC_QHL_CYCLES_N Counts uclk cycles all entries in the
OT_EMPTY.LOCAL
Quickpath Home Logic local tracker is
busy.
23H
01H
UNC_QHL_OCCUPAN QHL IOH tracker allocate to deallocate
CY.IOH
read occupancy.
23H
02H
UNC_QHL_OCCUPAN QHL remote tracker allocate to
CY.REMOTE
deallocate read occupancy.
23H
04H
UNC_QHL_OCCUPAN QHL local tracker allocate to
CY.LOCAL
deallocate read occupancy.
24H
02H
UNC_QHL_ADDRESS
_CONFLICTS.2WAY
Counts number of QHL Active Address
Table (AAT) entries that saw a max of
2 conflicts. The AAT is a structure that
tracks requests that are in conflict.
The requests themselves are in the
home tracker entries. The count is
reported when an AAT entry
deallocates.
24H
04H
UNC_QHL_ADDRESS
_CONFLICTS.3WAY
Counts number of QHL Active Address
Table (AAT) entries that saw a max of
3 conflicts. The AAT is a structure that
tracks requests that are in conflict.
The requests themselves are in the
home tracker entries. The count is
reported when an AAT entry
deallocates.
Description
Comment
Vol. 3B A-55
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
25H
01H
UNC_QHL_CONFLICT Counts cycles the Quickpath Home
_CYCLES.IOH
Logic IOH Tracker contains two or
more requests with an address
conflict. A max of 3 requests can be in
conflict.
25H
02H
UNC_QHL_CONFLICT Counts cycles the Quickpath Home
_CYCLES.REMOTE
Logic Remote Tracker contains two or
more requests with an address
conflict. A max of 3 requests can be in
conflict.
25H
04H
UNC_QHL_CONFLICT Counts cycles the Quickpath Home
_CYCLES.LOCAL
Logic Local Tracker contains two or
more requests with an address
conflict. A max of 3 requests can be
in conflict.
26H
01H
UNC_QHL_TO_QMC_ Counts number or requests to the
BYPASS
Quickpath Memory Controller that
bypass the Quickpath Home Logic. All
local accesses can be bypassed. For
remote requests, only read requests
can be bypassed.
27H
01H
UNC_QMC_NORMAL_ Uncore cycles all the entries in the
FULL.READ.CH0
DRAM channel 0 medium or low
priority queue are occupied with read
requests.
27H
02H
UNC_QMC_NORMAL_ Uncore cycles all the entries in the
FULL.READ.CH1
DRAM channel 1 medium or low
priority queue are occupied with read
requests.
27H
04H
UNC_QMC_NORMAL_ Uncore cycles all the entries in the
FULL.READ.CH2
DRAM channel 2 medium or low
priority queue are occupied with read
requests.
27H
08H
UNC_QMC_NORMAL_ Uncore cycles all the entries in the
FULL.WRITE.CH0
DRAM channel 0 medium or low
priority queue are occupied with write
requests.
A-56 Vol. 3B
Description
Comment
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
27H
10H
UNC_QMC_NORMAL_ Counts cycles all the entries in the
FULL.WRITE.CH1
DRAM channel 1 medium or low
priority queue are occupied with write
requests.
27H
20H
UNC_QMC_NORMAL_ Uncore cycles all the entries in the
FULL.WRITE.CH2
DRAM channel 2 medium or low
priority queue are occupied with write
requests.
28H
01H
UNC_QMC_ISOC_FUL Counts cycles all the entries in the
L.READ.CH0
DRAM channel 0 high priority queue
are occupied with isochronous read
requests.
28H
02H
UNC_QMC_ISOC_FUL Counts cycles all the entries in the
L.READ.CH1
DRAM channel 1high priority queue
are occupied with isochronous read
requests.
28H
04H
UNC_QMC_ISOC_FUL Counts cycles all the entries in the
L.READ.CH2
DRAM channel 2 high priority queue
are occupied with isochronous read
requests.
28H
08H
UNC_QMC_ISOC_FUL Counts cycles all the entries in the
L.WRITE.CH0
DRAM channel 0 high priority queue
are occupied with isochronous write
requests.
28H
10H
UNC_QMC_ISOC_FUL Counts cycles all the entries in the
L.WRITE.CH1
DRAM channel 1 high priority queue
are occupied with isochronous write
requests.
28H
20H
UNC_QMC_ISOC_FUL Counts cycles all the entries in the
L.WRITE.CH2
DRAM channel 2 high priority queue
are occupied with isochronous write
requests.
29H
01H
UNC_QMC_BUSY.REA Counts cycles where Quickpath
D.CH0
Memory Controller has at least 1
outstanding read request to DRAM
channel 0.
Description
Comment
Vol. 3B A-57
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
29H
02H
UNC_QMC_BUSY.REA Counts cycles where Quickpath
D.CH1
Memory Controller has at least 1
outstanding read request to DRAM
channel 1.
29H
04H
UNC_QMC_BUSY.REA Counts cycles where Quickpath
D.CH2
Memory Controller has at least 1
outstanding read request to DRAM
channel 2.
29H
08H
UNC_QMC_BUSY.WRI Counts cycles where Quickpath
TE.CH0
Memory Controller has at least 1
outstanding write request to DRAM
channel 0.
29H
10H
UNC_QMC_BUSY.WRI Counts cycles where Quickpath
TE.CH1
Memory Controller has at least 1
outstanding write request to DRAM
channel 1.
29H
20H
UNC_QMC_BUSY.WRI Counts cycles where Quickpath
TE.CH2
Memory Controller has at least 1
outstanding write request to DRAM
channel 2.
2AH
01H
UNC_QMC_OCCUPAN IMC channel 0 normal read request
CY.CH0
occupancy.
2AH
02H
UNC_QMC_OCCUPAN IMC channel 1 normal read request
CY.CH1
occupancy.
2AH
04H
UNC_QMC_OCCUPAN IMC channel 2 normal read request
CY.CH2
occupancy.
2BH
01H
UNC_QMC_ISSOC_OC IMC channel 0 issoc read request
CUPANCY.CH0
occupancy.
2BH
02H
UNC_QMC_ISSOC_OC IMC channel 1 issoc read request
CUPANCY.CH1
occupancy.
2BH
04H
UNC_QMC_ISSOC_OC IMC channel 2 issoc read request
CUPANCY.CH2
occupancy.
2BH
07H
UNC_QMC_ISSOC_RE IMC issoc read request occupancy.
ADS.ANY
A-58 Vol. 3B
Description
Comment
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
2CH
01H
UNC_QMC_NORMAL_ Counts the number of Quickpath
READS.CH0
Memory Controller channel 0 medium
and low priority read requests. The
QMC channel 0 normal read
occupancy divided by this count
provides the average QMC channel 0
read latency.
2CH
02H
UNC_QMC_NORMAL_ Counts the number of Quickpath
READS.CH1
Memory Controller channel 1 medium
and low priority read requests. The
QMC channel 1 normal read
occupancy divided by this count
provides the average QMC channel 1
read latency.
2CH
04H
UNC_QMC_NORMAL_ Counts the number of Quickpath
READS.CH2
Memory Controller channel 2 medium
and low priority read requests. The
QMC channel 2 normal read
occupancy divided by this count
provides the average QMC channel 2
read latency.
2CH
07H
UNC_QMC_NORMAL_ Counts the number of Quickpath
READS.ANY
Memory Controller medium and low
priority read requests. The QMC
normal read occupancy divided by this
count provides the average QMC read
latency.
2DH
01H
UNC_QMC_HIGH_PRI Counts the number of Quickpath
ORITY_READS.CH0
Memory Controller channel 0 high
priority isochronous read requests.
2DH
02H
UNC_QMC_HIGH_PRI Counts the number of Quickpath
ORITY_READS.CH1
Memory Controller channel 1 high
priority isochronous read requests.
2DH
04H
UNC_QMC_HIGH_PRI Counts the number of Quickpath
ORITY_READS.CH2
Memory Controller channel 2 high
priority isochronous read requests.
2DH
07H
UNC_QMC_HIGH_PRI Counts the number of Quickpath
ORITY_READS.ANY
Memory Controller high priority
isochronous read requests.
Description
Comment
Vol. 3B A-59
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
2EH
01H
UNC_QMC_CRITICAL_ Counts the number of Quickpath
PRIORITY_READS.CH Memory Controller channel 0 critical
0
priority isochronous read requests.
2EH
02H
UNC_QMC_CRITICAL_ Counts the number of Quickpath
PRIORITY_READS.CH Memory Controller channel 1 critical
1
priority isochronous read requests.
2EH
04H
UNC_QMC_CRITICAL_ Counts the number of Quickpath
PRIORITY_READS.CH Memory Controller channel 2 critical
2
priority isochronous read requests.
2EH
07H
UNC_QMC_CRITICAL_ Counts the number of Quickpath
PRIORITY_READS.AN Memory Controller critical priority
isochronous read requests.
Y
2FH
01H
UNC_QMC_WRITES.F
ULL.CH0
Counts number of full cache line
writes to DRAM channel 0.
2FH
02H
UNC_QMC_WRITES.F
ULL.CH1
Counts number of full cache line
writes to DRAM channel 1.
2FH
04H
UNC_QMC_WRITES.F
ULL.CH2
Counts number of full cache line
writes to DRAM channel 2.
2FH
07H
UNC_QMC_WRITES.F
ULL.ANY
Counts number of full cache line
writes to DRAM.
2FH
08H
UNC_QMC_WRITES.P Counts number of partial cache line
ARTIAL.CH0
writes to DRAM channel 0.
2FH
10H
UNC_QMC_WRITES.P Counts number of partial cache line
ARTIAL.CH1
writes to DRAM channel 1.
2FH
20H
UNC_QMC_WRITES.P Counts number of partial cache line
ARTIAL.CH2
writes to DRAM channel 2.
2FH
38H
UNC_QMC_WRITES.P Counts number of partial cache line
ARTIAL.ANY
writes to DRAM.
30H
01H
UNC_QMC_CANCEL.C Counts number of DRAM channel 0
H0
cancel requests.
30H
02H
UNC_QMC_CANCEL.C Counts number of DRAM channel 1
H1
cancel requests.
30H
04H
UNC_QMC_CANCEL.C Counts number of DRAM channel 2
H2
cancel requests.
30H
07H
UNC_QMC_CANCEL.A Counts number of DRAM cancel
NY
requests.
A-60 Vol. 3B
Description
Comment
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
31H
01H
UNC_QMC_PRIORITY Counts number of DRAM channel 0
_UPDATES.CH0
priority updates. A priority update
occurs when an ISOC high or critical
request is received by the QHL and
there is a matching request with
normal priority that has already been
issued to the QMC. In this instance,
the QHL will send a priority update to
QMC to expedite the request.
31H
02H
UNC_QMC_PRIORITY Counts number of DRAM channel 1
_UPDATES.CH1
priority updates. A priority update
occurs when an ISOC high or critical
request is received by the QHL and
there is a matching request with
normal priority that has already been
issued to the QMC. In this instance,
the QHL will send a priority update to
QMC to expedite the request.
31H
04H
UNC_QMC_PRIORITY Counts number of DRAM channel 2
_UPDATES.CH2
priority updates. A priority update
occurs when an ISOC high or critical
request is received by the QHL and
there is a matching request with
normal priority that has already been
issued to the QMC. In this instance,
the QHL will send a priority update to
QMC to expedite the request.
31H
07H
UNC_QMC_PRIORITY Counts number of DRAM priority
_UPDATES.ANY
updates. A priority update occurs
when an ISOC high or critical request
is received by the QHL and there is a
matching request with normal priority
that has already been issued to the
QMC. In this instance, the QHL will
send a priority update to QMC to
expedite the request.
33H
04H
UNC_QHL_FRC_ACK_ Counts number of Force Acknowledge
CNFLTS.LOCAL
Conflict messages sent by the
Quickpath Home Logic to the local
home.
Description
Comment
Vol. 3B A-61
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
40H
01H
UNC_QPI_TX_STALL
ED_SINGLE_FLIT.HO
ME.LINK_0
40H
02H
UNC_QPI_TX_STALL Counts cycles the Quickpath outbound
ED_SINGLE_FLIT.SNO link 0 SNOOP virtual channel is stalled
OP.LINK_0
due to lack of a VNA and VN0 credit.
Note that this event does not filter
out when a flit would not have been
selected for arbitration because
another virtual channel is getting
arbitrated.
40H
04H
UNC_QPI_TX_STALL Counts cycles the Quickpath outbound
ED_SINGLE_FLIT.NDR link 0 non-data response virtual
.LINK_0
channel is stalled due to lack of a VNA
and VN0 credit. Note that this event
does not filter out when a flit would
not have been selected for arbitration
because another virtual channel is
getting arbitrated.
40H
08H
UNC_QPI_TX_STALL
ED_SINGLE_FLIT.HO
ME.LINK_1
40H
10H
UNC_QPI_TX_STALL Counts cycles the Quickpath outbound
ED_SINGLE_FLIT.SNO link 1 SNOOP virtual channel is stalled
OP.LINK_1
due to lack of a VNA and VN0 credit.
Note that this event does not filter
out when a flit would not have been
selected for arbitration because
another virtual channel is getting
arbitrated.
A-62 Vol. 3B
Description
Counts cycles the Quickpath outbound
link 0 HOME virtual channel is stalled
due to lack of a VNA and VN0 credit.
Note that this event does not filter
out when a flit would not have been
selected for arbitration because
another virtual channel is getting
arbitrated.
Counts cycles the Quickpath outbound
link 1 HOME virtual channel is stalled
due to lack of a VNA and VN0 credit.
Note that this event does not filter
out when a flit would not have been
selected for arbitration because
another virtual channel is getting
arbitrated.
Comment
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
40H
20H
UNC_QPI_TX_STALL Counts cycles the Quickpath outbound
ED_SINGLE_FLIT.NDR link 1 non-data response virtual
.LINK_1
channel is stalled due to lack of a VNA
and VN0 credit. Note that this event
does not filter out when a flit would
not have been selected for arbitration
because another virtual channel is
getting arbitrated.
40H
07H
UNC_QPI_TX_STALL
ED_SINGLE_FLIT.LIN
K_0
Counts cycles the Quickpath outbound
link 0 virtual channels are stalled due
to lack of a VNA and VN0 credit. Note
that this event does not filter out
when a flit would not have been
selected for arbitration because
another virtual channel is getting
arbitrated.
40H
38H
UNC_QPI_TX_STALL
ED_SINGLE_FLIT.LIN
K_1
Counts cycles the Quickpath outbound
link 1 virtual channels are stalled due
to lack of a VNA and VN0 credit. Note
that this event does not filter out
when a flit would not have been
selected for arbitration because
another virtual channel is getting
arbitrated.
41H
01H
UNC_QPI_TX_STALL
ED_MULTI_FLIT.DRS.
LINK_0
Counts cycles the Quickpath outbound
link 0 Data ResponSe virtual channel
is stalled due to lack of VNA and VN0
credits. Note that this event does not
filter out when a flit would not have
been selected for arbitration because
another virtual channel is getting
arbitrated.
41H
02H
UNC_QPI_TX_STALL
ED_MULTI_FLIT.NCB.
LINK_0
Counts cycles the Quickpath outbound
link 0 Non-Coherent Bypass virtual
channel is stalled due to lack of VNA
and VN0 credits. Note that this event
does not filter out when a flit would
not have been selected for arbitration
because another virtual channel is
getting arbitrated.
Description
Comment
Vol. 3B A-63
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
41H
04H
UNC_QPI_TX_STALL
ED_MULTI_FLIT.NCS.
LINK_0
Counts cycles the Quickpath outbound
link 0 Non-Coherent Standard virtual
channel is stalled due to lack of VNA
and VN0 credits. Note that this event
does not filter out when a flit would
not have been selected for arbitration
because another virtual channel is
getting arbitrated.
41H
08H
UNC_QPI_TX_STALL
ED_MULTI_FLIT.DRS.
LINK_1
Counts cycles the Quickpath outbound
link 1 Data ResponSe virtual channel
is stalled due to lack of VNA and VN0
credits. Note that this event does not
filter out when a flit would not have
been selected for arbitration because
another virtual channel is getting
arbitrated.
41H
10H
UNC_QPI_TX_STALL
ED_MULTI_FLIT.NCB.
LINK_1
Counts cycles the Quickpath outbound
link 1 Non-Coherent Bypass virtual
channel is stalled due to lack of VNA
and VN0 credits. Note that this event
does not filter out when a flit would
not have been selected for arbitration
because another virtual channel is
getting arbitrated.
41H
20H
UNC_QPI_TX_STALL
ED_MULTI_FLIT.NCS.
LINK_1
Counts cycles the Quickpath outbound
link 1 Non-Coherent Standard virtual
channel is stalled due to lack of VNA
and VN0 credits. Note that this event
does not filter out when a flit would
not have been selected for arbitration
because another virtual channel is
getting arbitrated.
41H
07H
UNC_QPI_TX_STALL
ED_MULTI_FLIT.LINK
_0
Counts cycles the Quickpath outbound
link 0 virtual channels are stalled due
to lack of VNA and VN0 credits. Note
that this event does not filter out
when a flit would not have been
selected for arbitration because
another virtual channel is getting
arbitrated.
A-64 Vol. 3B
Description
Comment
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
41H
38H
UNC_QPI_TX_STALL
ED_MULTI_FLIT.LINK
_1
42H
02H
UNC_QPI_TX_HEADE Number of cycles that the header
R.BUSY.LINK_0
buffer in the Quickpath Interface
outbound link 0 is busy.
42H
08H
UNC_QPI_TX_HEADE Number of cycles that the header
R.BUSY.LINK_1
buffer in the Quickpath Interface
outbound link 1 is busy.
43H
01H
UNC_QPI_RX_NO_PP Number of cycles that snoop packets
T_CREDIT.STALLS.LIN incoming to the Quickpath Interface
K_0
link 0 are stalled and not sent to the
GQ because the GQ Peer Probe
Tracker (PPT) does not have any
available entries.
43H
02H
UNC_QPI_RX_NO_PP Number of cycles that snoop packets
T_CREDIT.STALLS.LIN incoming to the Quickpath Interface
K_1
link 1 are stalled and not sent to the
GQ because the GQ Peer Probe
Tracker (PPT) does not have any
available entries.
60H
01H
UNC_DRAM_OPEN.C
H0
Counts number of DRAM Channel 0
open commands issued either for read
or write. To read or write data, the
referenced DRAM page must first be
opened.
60H
02H
UNC_DRAM_OPEN.C
H1
Counts number of DRAM Channel 1
open commands issued either for read
or write. To read or write data, the
referenced DRAM page must first be
opened.
Description
Comment
Counts cycles the Quickpath outbound
link 1 virtual channels are stalled due
to lack of VNA and VN0 credits. Note
that this event does not filter out
when a flit would not have been
selected for arbitration because
another virtual channel is getting
arbitrated.
Vol. 3B A-65
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
60H
04H
UNC_DRAM_OPEN.C
H2
Counts number of DRAM Channel 2
open commands issued either for read
or write. To read or write data, the
referenced DRAM page must first be
opened.
61H
01H
UNC_DRAM_PAGE_C
LOSE.CH0
DRAM channel 0 command issued to
CLOSE a page due to page idle timer
expiration. Closing a page is done by
issuing a precharge.
61H
02H
UNC_DRAM_PAGE_C
LOSE.CH1
DRAM channel 1 command issued to
CLOSE a page due to page idle timer
expiration. Closing a page is done by
issuing a precharge.
61H
04H
UNC_DRAM_PAGE_C
LOSE.CH2
DRAM channel 2 command issued to
CLOSE a page due to page idle timer
expiration. Closing a page is done by
issuing a precharge.
62H
01H
UNC_DRAM_PAGE_M Counts the number of precharges
ISS.CH0
(PRE) that were issued to DRAM
channel 0 because there was a page
miss. A page miss refers to a situation
in which a page is currently open and
another page from the same bank
needs to be opened. The new page
experiences a page miss. Closing of
the old page is done by issuing a
precharge.
62H
02H
UNC_DRAM_PAGE_M Counts the number of precharges
ISS.CH1
(PRE) that were issued to DRAM
channel 1 because there was a page
miss. A page miss refers to a situation
in which a page is currently open and
another page from the same bank
needs to be opened. The new page
experiences a page miss. Closing of
the old page is done by issuing a
precharge.
A-66 Vol. 3B
Description
Comment
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
62H
04H
UNC_DRAM_PAGE_M Counts the number of precharges
ISS.CH2
(PRE) that were issued to DRAM
channel 2 because there was a page
miss. A page miss refers to a situation
in which a page is currently open and
another page from the same bank
needs to be opened. The new page
experiences a page miss. Closing of
the old page is done by issuing a
precharge.
63H
01H
UNC_DRAM_READ_C Counts the number of times a read
AS.CH0
CAS command was issued on DRAM
channel 0.
63H
02H
UNC_DRAM_READ_C Counts the number of times a read
AS.AUTOPRE_CH0
CAS command was issued on DRAM
channel 0 where the command issued
used the auto-precharge (auto page
close) mode.
63H
04H
UNC_DRAM_READ_C Counts the number of times a read
AS.CH1
CAS command was issued on DRAM
channel 1.
63H
08H
UNC_DRAM_READ_C Counts the number of times a read
AS.AUTOPRE_CH1
CAS command was issued on DRAM
channel 1 where the command issued
used the auto-precharge (auto page
close) mode.
63H
10H
UNC_DRAM_READ_C Counts the number of times a read
AS.CH2
CAS command was issued on DRAM
channel 2.
63H
20H
UNC_DRAM_READ_C Counts the number of times a read
AS.AUTOPRE_CH2
CAS command was issued on DRAM
channel 2 where the command issued
used the auto-precharge (auto page
close) mode.
64H
01H
UNC_DRAM_WRITE_
CAS.CH0
Description
Comment
Counts the number of times a write
CAS command was issued on DRAM
channel 0.
Vol. 3B A-67
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
64H
02H
UNC_DRAM_WRITE_
CAS.AUTOPRE_CH0
Counts the number of times a write
CAS command was issued on DRAM
channel 0 where the command issued
used the auto-precharge (auto page
close) mode.
64H
04H
UNC_DRAM_WRITE_
CAS.CH1
Counts the number of times a write
CAS command was issued on DRAM
channel 1.
64H
08H
UNC_DRAM_WRITE_
CAS.AUTOPRE_CH1
Counts the number of times a write
CAS command was issued on DRAM
channel 1 where the command issued
used the auto-precharge (auto page
close) mode.
64H
10H
UNC_DRAM_WRITE_
CAS.CH2
Counts the number of times a write
CAS command was issued on DRAM
channel 2.
64H
20H
UNC_DRAM_WRITE_
CAS.AUTOPRE_CH2
Counts the number of times a write
CAS command was issued on DRAM
channel 2 where the command issued
used the auto-precharge (auto page
close) mode.
65H
01H
UNC_DRAM_REFRES Counts number of DRAM channel 0
H.CH0
refresh commands. DRAM loses data
content over time. In order to keep
correct data content, the data values
have to be refreshed periodically.
65H
02H
UNC_DRAM_REFRES Counts number of DRAM channel 1
H.CH1
refresh commands. DRAM loses data
content over time. In order to keep
correct data content, the data values
have to be refreshed periodically.
65H
04H
UNC_DRAM_REFRES Counts number of DRAM channel 2
H.CH2
refresh commands. DRAM loses data
content over time. In order to keep
correct data content, the data values
have to be refreshed periodically.
A-68 Vol. 3B
Description
Comment
PERFORMANCE-MONITORING EVENTS
Table A-5. Non-Architectural Performance Events In the Processor Uncore for Intel
Core i7 Processor and Intel Xeon Processor 5500 Series
Event
Num.
Umask
Value
Event Mask
Mnemonic
66H
01H
UNC_DRAM_PRE_AL Counts number of DRAM Channel 0
L.CH0
precharge-all (PREALL) commands
that close all open pages in a rank.
PREALL is issued when the DRAM
needs to be refreshed or needs to go
into a power down mode.
66H
02H
UNC_DRAM_PRE_AL Counts number of DRAM Channel 1
L.CH1
precharge-all (PREALL) commands
that close all open pages in a rank.
PREALL is issued when the DRAM
needs to be refreshed or needs to go
into a power down mode.
66H
04H
UNC_DRAM_PRE_AL Counts number of DRAM Channel 2
L.CH2
precharge-all (PREALL) commands
that close all open pages in a rank.
PREALL is issued when the DRAM
needs to be refreshed or needs to go
into a power down mode.
Description
Comment
Intel Xeon processors with CPUID signature of DisplayFamily_DisplayModel 06_2EH
have a distinct uncore sub-system that is significantly different from the uncore
found in processors with CPUID signature 06_1AH, 06_1EH, and 06_1FH. Non-architectural Performance monitoring events for its uncore will be available in future documentation.
A.4
PERFORMANCE MONITORING EVENTS FOR
PROCESSORS BASED ON
INTEL® MICROARCHITECTURE CODE NAME
WESTMERE
Intel 64 processors based on Intel® microarchitecture code name Westmere support
the architectural and non-architectural performance-monitoring events listed in
Table A-1 and Table A-6. Table A-6 applies to processors with CPUID signature of
DisplayFamily_DisplayModel encoding with the following values: 06_25H, 06_2CH.
In addition, these processors (CPUID signature of DisplayFamily_DisplayModel
06_25H, 06_2CH) also support the following non-architectural, product-specific
uncore performance-monitoring events listed in Table A-7. Fixed counters support
the architecture events defined in Table A-9.
Vol. 3B A-69
PERFORMANCE-MONITORING EVENTS
Table A-6. Non-Architectural Performance Events In the Processor Core for Processors
Based on Intel Microarchitecture Code Name Westmere
Event
Num.
Umask
Value
Event Mask
Mnemonic
Description
03H
02H
LOAD_BLOCK.OVERL
AP_STORE
Loads that partially overlap an
earlier store.
04H
07H
SB_DRAIN.ANY
All Store buffer stall cycles.
05H
02H
MISALIGN_MEMORY.S All store referenced with misaligned
TORE
address.
06H
04H
STORE_BLOCKS.AT_
RET
Counts number of loads delayed
with at-Retirement block code. The
following loads need to be executed
at retirement and wait for all senior
stores on the same thread to be
drained: load splitting across 4K
boundary (page split), load accessing
uncacheable (UC or USWC) memory,
load lock, and load with page table in
UC or USWC memory region.
06H
08H
STORE_BLOCKS.L1D
_BLOCK
Cacheable loads delayed with L1D
block code.
07H
01H
PARTIAL_ADDRESS_
ALIAS
Counts false dependency due to
partial address aliasing.
08H
01H
DTLB_LOAD_MISSES. Counts all load misses that cause a
ANY
page walk.
08H
02H
DTLB_LOAD_MISSES. Counts number of completed page
WALK_COMPLETED
walks due to load miss in the STLB.
08H
04H
DTLB_LOAD_MISSES. Cycles PMH is busy with a page walk
WALK_CYCLES
due to a load miss in the STLB.
08H
10H
DTLB_LOAD_MISSES. Number of cache load STLB hits.
STLB_HIT
08H
20H
DTLB_LOAD_MISSES. Number of DTLB cache load misses
PDE_MISS
where the low part of the linear to
physical address translation was
missed.
0BH
01H
MEM_INST_RETIRED. Counts the number of instructions
LOADS
with an architecturally-visible load
retired on the architected path.
0BH
02H
MEM_INST_RETIRED. Counts the number of instructions
STORES
with an architecturally-visible store
retired on the architected path.
A-70 Vol. 3B
Comment
PERFORMANCE-MONITORING EVENTS
Table A-6. Non-Architectural Performance Events In the Processor Core for Processors
Based on Intel Microarchitecture Code Name Westmere
Event
Num.
Umask
Value
Event Mask
Mnemonic
0BH
10H
MEM_INST_RETIRED. Counts the number of instructions
In conjunction
LATENCY_ABOVE_T exceeding the latency specified with with ld_lat
HRESHOLD
ld_lat facility.
facility
0CH
01H
MEM_STORE_RETIRE The event counts the number of
D.DTLB_MISS
retired stores that missed the DTLB.
The DTLB miss is not counted if the
store operation causes a fault. Does
not counter prefetches. Counts both
primary and secondary misses to
the TLB.
0EH
01H
UOPS_ISSUED.ANY
0EH
01H
UOPS_ISSUED.STALL Counts the number of cycles no
ED_CYCLES
Uops issued by the Register
Allocation Table to the Reservation
Station, i.e. the UOPs issued from
the front end to the back end.
0EH
02H
UOPS_ISSUED.FUSED Counts the number of fused Uops
that were issued from the Register
Allocation Table to the Reservation
Station.
0FH
01H
MEM_UNCORE_RETI Load instructions retired with
RED.UNKNOWN_SOU unknown LLC miss (Precise Event).
RCE
0FH
02H
MEM_UNCORE_RETI
RED.OHTER_CORE_L
2_HIT
Load instructions retired that HIT
Applicable to one
modified data in sibling core (Precise and two sockets
Event).
0FH
04H
MEM_UNCORE_RETI
RED.REMOTE_HITM
Load instructions retired that HIT
modified data in remote socket
(Precise Event).
0FH
08H
MEM_UNCORE_RETI Load instructions retired local dram
RED.LOCAL_DRAM_A and remote cache HIT data sources
ND_REMOTE_CACHE (Precise Event).
_HIT
Description
Comment
Counts the number of Uops issued
by the Register Allocation Table to
the Reservation Station, i.e. the
UOPs issued from the front end to
the back end.
set “invert=1,
cmask = 1“
Applicable to one
and two sockets
Applicable to two
sockets only
Applicable to one
and two sockets
Vol. 3B A-71
PERFORMANCE-MONITORING EVENTS
Table A-6. Non-Architectural Performance Events In the Processor Core for Processors
Based on Intel Microarchitecture Code Name Westmere
Event
Num.
Umask
Value
Event Mask
Mnemonic
0FH
10H
MEM_UNCORE_RETI
RED.REMOTE_DRAM
0FH
20H
MEM_UNCORE_RETI Load instructions retired other LLC
RED.OTHER_LLC_MIS miss (Precise Event).
S
0FH
80H
MEM_UNCORE_RETI
RED.UNCACHEABLE
Load instructions retired I/O (Precise Applicable to one
Event).
and two sockets
10H
01H
FP_COMP_OPS_EXE.
X87
Counts the number of FP
Computational Uops Executed. The
number of FADD, FSUB, FCOM,
FMULs, integer MULsand IMULs,
FDIVs, FPREMs, FSQRTS, integer
DIVs, and IDIVs. This event does not
distinguish an FADD used in the
middle of a transcendental flow
from a separate FADD instruction.
10H
02H
FP_COMP_OPS_EXE.
MMX
Counts number of MMX Uops
executed.
10H
04H
FP_COMP_OPS_EXE.
SSE_FP
Counts number of SSE and SSE2 FP
uops executed.
10H
08H
FP_COMP_OPS_EXE.
SSE2_INTEGER
Counts number of SSE2 integer uops
executed.
10H
10H
FP_COMP_OPS_EXE.
SSE_FP_PACKED
Counts number of SSE FP packed
uops executed.
10H
20H
FP_COMP_OPS_EXE.
SSE_FP_SCALAR
Counts number of SSE FP scalar
uops executed.
10H
40H
FP_COMP_OPS_EXE. Counts number of SSE* FP single
SSE_SINGLE_PRECISI precision uops executed.
ON
10H
80H
FP_COMP_OPS_EXE.
SSE_DOUBLE_PRECI
SION
12H
01H
SIMD_INT_128.PACK Counts number of 128 bit SIMD
ED_MPY
integer multiply operations.
12H
02H
SIMD_INT_128.PACK Counts number of 128 bit SIMD
ED_SHIFT
integer shift operations.
A-72 Vol. 3B
Description
Comment
Load instructions retired remote
DRAM and remote home-remote
cache HITM (Precise Event).
Applicable to two
sockets only
Counts number of SSE* FP double
precision uops executed.
Applicable to two
sockets only
PERFORMANCE-MONITORING EVENTS
Table A-6. Non-Architectural Performance Events In the Processor Core for Processors
Based on Intel Microarchitecture Code Name Westmere
Event
Num.
Umask
Value
Event Mask
Mnemonic
12H
04H
SIMD_INT_128.PACK Counts number of 128 bit SIMD
integer pack operations.
12H
08H
SIMD_INT_128.UNPA Counts number of 128 bit SIMD
CK
integer unpack operations.
12H
10H
SIMD_INT_128.PACK Counts number of 128 bit SIMD
ED_LOGICAL
integer logical operations.
12H
20H
SIMD_INT_128.PACK Counts number of 128 bit SIMD
ED_ARITH
integer arithmetic operations.
12H
40H
SIMD_INT_128.SHUF Counts number of 128 bit SIMD
FLE_MOVE
integer shuffle and move
operations.
13H
01H
LOAD_DISPATCH.RS
13H
02H
LOAD_DISPATCH.RS_ Counts the number of delayed RS
DELAYED
dispatches at the stage latch. If an
RS dispatch can not bypass to LB, it
has another chance to dispatch from
the one-cycle delayed staging latch
before it is written into the LB.
13H
04H
LOAD_DISPATCH.MO
B
13H
07H
LOAD_DISPATCH.ANY Counts all loads dispatched from the
Reservation Station.
14H
01H
ARITH.CYCLES_DIV_
BUSY
Description
Comment
Counts number of loads dispatched
from the Reservation Station that
bypass the Memory Order Buffer.
Counts the number of loads
dispatched from the Reservation
Station to the Memory Order Buffer.
Counts the number of cycles the
divider is busy executing divide or
square root operations. The divide
can be integer, X87 or Streaming
SIMD Extensions (SSE). The square
root operation can be either X87 or
SSE.
Count may be
incorrect When
SMT is on
Set 'edge =1, invert=1, cmask=1' to
count the number of divides.
Vol. 3B A-73
PERFORMANCE-MONITORING EVENTS
Table A-6. Non-Architectural Performance Events In the Processor Core for Processors
Based on Intel Microarchitecture Code Name Westmere
Event
Num.
Umask
Value
Event Mask
Mnemonic
14H
02H
ARITH.MUL
17H
01H
INST_QUEUE_WRITE Counts the number of instructions
S
written into the instruction queue
every cycle.
18H
01H
INST_DECODED.DEC0 Counts number of instructions that
require decoder 0 to be decoded.
Usually, this means that the
instruction maps to more than 1
uop.
19H
01H
TWO_UOP_INSTS_D
ECODED
1EH
01H
INST_QUEUE_WRITE This event counts the number of
_CYCLES
cycles during which instructions are
written to the instruction queue.
Dividing this counter by the number
of instructions written to the
instruction queue
(INST_QUEUE_WRITES) yields the
average number of instructions
decoded each cycle. If this number is
less than four and the pipe stalls,
this indicates that the decoder is
failing to decode enough
instructions per cycle to sustain the
4-wide pipeline.
20H
01H
LSD_OVERFLOW
Number of loops that can not stream
from the instruction queue.
24H
01H
L2_RQSTS.LD_HIT
Counts number of loads that hit the
L2 cache. L2 loads include both L1D
demand misses as well as L1D
prefetches. L2 loads can be rejected
for various reasons. Only non
rejected loads are counted.
A-74 Vol. 3B
Description
Comment
Counts the number of multiply
operations executed. This includes
integer as well as floating point
multiply operations but excludes
DPPS mul and MPSAD.
Count may be
incorrect When
SMT is on
An instruction that generates two
uops was decoded.
If SSE*
instructions that
are 6 bytes or
longer arrive one
after another,
then front end
throughput may
limit execution
speed.
PERFORMANCE-MONITORING EVENTS
Table A-6. Non-Architectural Performance Events In the Processor Core for Processors
Based on Intel Microarchitecture Code Name Westmere
Event
Num.
Umask
Value
Event Mask
Mnemonic
24H
02H
L2_RQSTS.LD_MISS
Counts the number of loads that
miss the L2 cache. L2 loads include
both L1D demand misses as well as
L1D prefetches.
24H
03H
L2_RQSTS.LOADS
Counts all L2 load requests. L2 loads
include both L1D demand misses as
well as L1D prefetches.
24H
04H
L2_RQSTS.RFO_HIT
Counts the number of store RFO
requests that hit the L2 cache. L2
RFO requests include both L1D
demand RFO misses as well as L1D
RFO prefetches. Count includes WC
memory requests, where the data is
not fetched but the permission to
write the line is required.
24H
08H
L2_RQSTS.RFO_MISS Counts the number of store RFO
requests that miss the L2 cache. L2
RFO requests include both L1D
demand RFO misses as well as L1D
RFO prefetches.
24H
0CH
L2_RQSTS.RFOS
24H
10H
L2_RQSTS.IFETCH_H Counts number of instruction
IT
fetches that hit the L2 cache. L2
instruction fetches include both L1I
demand misses as well as L1I
instruction prefetches.
24H
20H
L2_RQSTS.IFETCH_M Counts number of instruction
ISS
fetches that miss the L2 cache. L2
instruction fetches include both L1I
demand misses as well as L1I
instruction prefetches.
24H
30H
L2_RQSTS.IFETCHES Counts all instruction fetches. L2
instruction fetches include both L1I
demand misses as well as L1I
instruction prefetches.
Description
Comment
Counts all L2 store RFO requests. L2
RFO requests include both L1D
demand RFO misses as well as L1D
RFO prefetches..
Vol. 3B A-75
PERFORMANCE-MONITORING EVENTS
Table A-6. Non-Architectural Performance Events In the Processor Core for Processors
Based on Intel Microarchitecture Code Name Westmere
Event
Num.
Umask
Value
Event Mask
Mnemonic
Description
24H
40H
L2_RQSTS.PREFETC
H_HIT
Counts L2 prefetch hits for both
code and data.
24H
80H
L2_RQSTS.PREFETC
H_MISS
Counts L2 prefetch misses for both
code and data.
24H
C0H
L2_RQSTS.PREFETC
HES
Counts all L2 prefetches for both
code and data.
24H
AAH
L2_RQSTS.MISS
Counts all L2 misses for both code
and data.
24H
FFH
L2_RQSTS.REFEREN
CES
Counts all L2 requests for both code
and data.
26H
01H
L2_DATA_RQSTS.DE
MAND.I_STATE
Counts number of L2 data demand
loads where the cache line to be
loaded is in the I (invalid) state, i.e. a
cache miss. L2 demand loads are
both L1D demand misses and L1D
prefetches.
26H
02H
L2_DATA_RQSTS.DE
MAND.S_STATE
Counts number of L2 data demand
loads where the cache line to be
loaded is in the S (shared) state. L2
demand loads are both L1D demand
misses and L1D prefetches.
26H
04H
L2_DATA_RQSTS.DE
MAND.E_STATE
Counts number of L2 data demand
loads where the cache line to be
loaded is in the E (exclusive) state.
L2 demand loads are both L1D
demand misses and L1D prefetches.
26H
08H
L2_DATA_RQSTS.DE
MAND.M_STATE
Counts number of L2 data demand
loads where the cache line to be
loaded is in the M (modified) state.
L2 demand loads are both L1D
demand misses and L1D prefetches.
26H
0FH
L2_DATA_RQSTS.DE
MAND.MESI
Counts all L2 data demand requests.
L2 demand loads are both L1D
demand misses and L1D prefetches.
26H
10H
L2_DATA_RQSTS.PR
EFETCH.I_STATE
Counts number of L2 prefetch data
loads where the cache line to be
loaded is in the I (invalid) state, i.e. a
cache miss.
A-76 Vol. 3B
Comment
PERFORMANCE-MONITORING EVENTS
Table A-6. Non-Architectural Performance Events In the Processor Core for Processors
Based on Intel Microarchitecture Code Name Westmere
Event
Num.
Umask
Value
Event Mask
Mnemonic
26H
20H
L2_DATA_RQSTS.PR
EFETCH.S_STATE
Counts number of L2 prefetch data
loads where the cache line to be
loaded is in the S (shared) state. A
prefetch RFO will miss on an S state
line, while a prefetch read will hit on
an S state line.
26H
40H
L2_DATA_RQSTS.PR
EFETCH.E_STATE
Counts number of L2 prefetch data
loads where the cache line to be
loaded is in the E (exclusive) state.
26H
80H
L2_DATA_RQSTS.PR
EFETCH.M_STATE
Counts number of L2 prefetch data
loads where the cache line to be
loaded is in the M (modified) state.
26H
F0H
L2_DATA_RQSTS.PR
EFETCH.MESI
Counts all L2 prefetch requests.
26H
FFH
L2_DATA_RQSTS.AN Counts all L2 data requests.
Y
27H
01H
L2_WRITE.RFO.I_STA Counts number of L2 demand store This is a demand
TE
RFO requests where the cache line RFO request
to be loaded is in the I (invalid) state,
i.e, a cache miss. The L1D prefetcher
does not issue a RFO prefetch.
27H
02H
L2_WRITE.RFO.S_ST
ATE
27H
08H
L2_WRITE.RFO.M_ST Counts number of L2 store RFO
This is a demand
ATE
requests where the cache line to be RFO request
loaded is in the M (modified) state.
The L1D prefetcher does not issue a
RFO prefetch.
27H
0EH
L2_WRITE.RFO.HIT
Counts number of L2 store RFO
This is a demand
requests where the cache line to be RFO request
loaded is in either the S, E or M
states. The L1D prefetcher does not
issue a RFO prefetch.
27H
0FH
L2_WRITE.RFO.MESI
Counts all L2 store RFO
requests.The L1D prefetcher does
not issue a RFO prefetch.
Description
Comment
This is a demand
Counts number of L2 store RFO
requests where the cache line to be RFO request
loaded is in the S (shared) state. The
L1D prefetcher does not issue a RFO
prefetch,.
This is a demand
RFO request
Vol. 3B A-77
PERFORMANCE-MONITORING EVENTS
Table A-6. Non-Architectural Performance Events In the Processor Core for Processors
Based on Intel Microarchitecture Code Name Westmere
Event
Num.
Umask
Value
Event Mask
Mnemonic
27H
10H
L2_WRITE.LOCK.I_ST Counts number of L2 demand lock
ATE
RFO requests where the cache line
to be loaded is in the I (invalid) state,
i.e. a cache miss.
27H
20H
L2_WRITE.LOCK.S_S
TATE
Counts number of L2 lock RFO
requests where the cache line to be
loaded is in the S (shared) state.
27H
40H
L2_WRITE.LOCK.E_S
TATE
Counts number of L2 demand lock
RFO requests where the cache line
to be loaded is in the E (exclusive)
state.
27H
80H
L2_WRITE.LOCK.M_S
TATE
Counts number of L2 demand lock
RFO requests where the cache line
to be loaded is in the M (modified)
state.
27H
E0H
L2_WRITE.LOCK.HIT
Counts number of L2 demand lock
RFO requests where the cache line
to be loaded is in either the S, E, or
M state.
27H
F0H
L2_WRITE.LOCK.MESI Counts all L2 demand lock RFO
requests.
28H
01H
L1D_WB_L2.I_STATE Counts number of L1 writebacks to
the L2 where the cache line to be
written is in the I (invalid) state, i.e. a
cache miss.
28H
02H
L1D_WB_L2.S_STAT
E
Counts number of L1 writebacks to
the L2 where the cache line to be
written is in the S state.
28H
04H
L1D_WB_L2.E_STAT
E
Counts number of L1 writebacks to
the L2 where the cache line to be
written is in the E (exclusive) state.
28H
08H
L1D_WB_L2.M_STAT Counts number of L1 writebacks to
E
the L2 where the cache line to be
written is in the M (modified) state.
28H
0FH
L1D_WB_L2.MESI
A-78 Vol. 3B
Description
Counts all L1 writebacks to the L2 .
Comment
PERFORMANCE-MONITORING EVENTS
Table A-6. Non-Architectural Performance Events In the Processor Core for Processors
Based on Intel Microarchitecture Code Name Westmere
Event
Num.
Umask
Value
Event Mask
Mnemonic
2EH
02H
L3_LAT_CACHE.REFE Counts uncore Last Level Cache
see Table A-1
RENCE
references. Because cache
hierarchy, cache sizes and other
implementation-specific
characteristics; value comparison to
estimate performance differences is
not recommended.
2EH
01H
L3_LAT_CACHE.MISS Counts uncore Last Level Cache
see Table A-1
misses. Because cache hierarchy,
cache sizes and other
implementation-specific
characteristics; value comparison to
estimate performance differences is
not recommended.
3CH
00H
CPU_CLK_UNHALTED Counts the number of thread cycles see Table A-1
.THREAD_P
while the thread is not in a halt
state. The thread enters the halt
state when it is running the HLT
instruction. The core frequency may
change from time to time due to
power or thermal throttling.
3CH
01H
CPU_CLK_UNHALTED Increments at the frequency of TSC
.REF_P
when not halted.
49H
01H
DTLB_MISSES.ANY
49H
02H
DTLB_MISSES.WALK_ Counts number of misses in the
COMPLETED
STLB which resulted in a completed
page walk.
49H
04H
DTLB_MISSES.WALK_ Counts cycles of page walk due to
CYCLES
misses in the STLB.
49H
10H
DTLB_MISSES.STLB_ Counts the number of DTLB first
HIT
level misses that hit in the second
level TLB. This event is only
relevant if the core contains multiple
DTLB levels.
49H
20H
DTLB_MISSES.PDE_M Number of DTLB misses caused by
ISS
low part of address, includes
references to 2M pages because 2M
pages do not use the PDE.
Description
Comment
see Table A-1
Counts the number of misses in the
STLB which causes a page walk.
Vol. 3B A-79
PERFORMANCE-MONITORING EVENTS
Table A-6. Non-Architectural Performance Events In the Processor Core for Processors
Based on Intel Microarchitecture Code Name Westmere
Event
Num.
Umask
Value
Event Mask
Mnemonic
49H
80H
DTLB_MISSES.LARGE Counts number of completed large
_WALK_COMPLETED page walks due to misses in the
STLB.
4CH
01H
LOAD_HIT_PRE
Counts load operations sent to the
L1 data cache while a previous SSE
prefetch instruction to the same
cache line has started prefetching
but has not yet finished.
4EH
01H
L1D_PREFETCH.REQ
UESTS
Counter 0, 1 only
Counts number of hardware
prefetch requests dispatched out of
the prefetch FIFO.
4EH
02H
L1D_PREFETCH.MISS Counts number of hardware
Counter 0, 1 only
prefetch requests that miss the L1D.
There are two prefetchers in the
L1D. A streamer, which predicts
lines sequentially after this one
should be fetched, and the IP
prefetcher that remembers access
patterns for the current instruction.
The streamer prefetcher stops on an
L1D hit, while the IP prefetcher
does not.
4EH
04H
L1D_PREFETCH.TRIG Counts number of prefetch requests Counter 0, 1 only
GERS
triggered by the Finite State
Machine and pushed into the
prefetch FIFO. Some of the prefetch
requests are dropped due to
overwrites or competition between
the IP index prefetcher and
streamer prefetcher. The prefetch
FIFO contains 4 entries.
4FH
10H
EPT.WALK_CYCLES
Counts Extended Page walk cycles.
51H
01H
L1D.REPL
Counts the number of lines brought
into the L1 data cache.
51H
02H
L1D.M_REPL
Counts the number of modified lines Counter 0, 1 only
brought into the L1 data cache.
51H
04H
L1D.M_EVICT
Counts the number of modified lines Counter 0, 1 only
evicted from the L1 data cache due
to replacement.
A-80 Vol. 3B
Description
Comment
Counter 0, 1 only
Counter 0, 1 only
PERFORMANCE-MONITORING EVENTS
Table A-6. Non-Architectural Performance Events In the Processor Core for Processors
Based on Intel Microarchitecture Code Name Westmere
Event
Num.
Umask
Value
Event Mask
Mnemonic
51H
08H
L1D.M_SNOOP_EVIC
T
52H
01H
L1D_CACHE_PREFET Counts the number of cacheable
CH_LOCK_FB_HIT
load lock speculated instructions
accepted into the fill buffer.
60H
01H
OFFCORE_REQUEST
S_OUTSTANDING.DE
MAND.READ_DATA
Counts weighted cycles of offcore
demand data read requests. Does
not include L2 prefetch requests.
counter 0
60H
02H
OFFCORE_REQUEST
S_OUTSTANDING.DE
MAND.READ_CODE
Counts weighted cycles of offcore
demand code read requests. Does
not include L2 prefetch requests.
counter 0
60H
04H
OFFCORE_REQUEST
S_OUTSTANDING.DE
MAND.RFO
Counts weighted cycles of offcore
demand RFO requests. Does not
include L2 prefetch requests.
counter 0
60H
08H
OFFCORE_REQUEST Counts weighted cycles of offcore
counter 0
S_OUTSTANDING.AN read requests of any kind. Include L2
Y.READ
prefetch requests.
63H
01H
CACHE_LOCK_CYCLE
S.L1D_L2
Cycle count during which the L1D
and L2 are locked. A lock is asserted
when there is a locked memory
access, due to uncacheable memory,
a locked operation that spans two
cache lines, or a page walk from an
uncacheable page table. This event
does not cause locks, it merely
detects them.
63H
02H
CACHE_LOCK_CYCLE
S.L1D
Counts the number of cycles that
Counter 0, 1 only.
cacheline in the L1 data cache unit is
locked.
6CH
01H
IO_TRANSACTIONS
Counts the number of completed I/O
transactions.
80H
01H
L1I.HITS
Counts all instruction fetches that
hit the L1 instruction cache.
Description
Comment
Counts the number of modified lines Counter 0, 1 only
evicted from the L1 data cache due
to snoop HITM intervention.
Counter 0, 1 only.
L1D and L2 locks
have a very high
performance
penalty and it is
highly
recommended to
avoid such
accesses.
Vol. 3B A-81
PERFORMANCE-MONITORING EVENTS
Table A-6. Non-Architectural Performance Events In the Processor Core for Processors
Based on Intel Microarchitecture Code Name Westmere
Event
Num.
Umask
Value
Event Mask
Mnemonic
80H
02H
L1I.MISSES
Counts all instruction fetches that
miss the L1I cache. This includes
instruction cache misses, streaming
buffer misses, victim cache misses
and uncacheable fetches. An
instruction fetch miss is counted
only once and not once for every
cycle it is outstanding.
80H
03H
L1I.READS
Counts all instruction fetches,
including uncacheable fetches that
bypass the L1I.
80H
04H
L1I.CYCLES_STALLED Cycle counts for which an instruction
fetch stalls due to a L1I cache miss,
ITLB miss or ITLB fault.
82H
01H
LARGE_ITLB.HIT
Counts number of large ITLB hits.
85H
01H
ITLB_MISSES.ANY
Counts the number of misses in all
levels of the ITLB which causes a
page walk.
85H
02H
ITLB_MISSES.WALK_
COMPLETED
Counts number of misses in all levels
of the ITLB which resulted in a
completed page walk.
85H
04H
ITLB_MISSES.WALK_
CYCLES
Counts ITLB miss page walk cycles.
85H
80H
ITLB_MISSES.LARGE_ Counts number of completed large
WALK_COMPLETED
page walks due to misses in the
STLB.
87H
01H
ILD_STALL.LCP
Cycles Instruction Length Decoder
stalls due to length changing
prefixes: 66, 67 or REX.W (for
EM64T) instructions which change
the length of the decoded
instruction.
87H
02H
ILD_STALL.MRU
Instruction Length Decoder stall
cycles due to Brand Prediction Unit
(PBU) Most Recently Used (MRU)
bypass.
87H
04H
ILD_STALL.IQ_FULL
Stall cycles due to a full instruction
queue.
A-82 Vol. 3B
Description
Comment
PERFORMANCE-MONITORING EVENTS
Table A-6. Non-Architectural Performance Events In the Processor Core for Processors
Based on Intel Microarchitecture Code Name Westmere
Event
Num.
Umask
Value
Event Mask
Mnemonic
Description
87H
08H
ILD_STALL.REGEN
Counts the number of regen stalls.
87H
0FH
ILD_STALL.ANY
Counts any cycles the Instruction
Length Decoder is stalled.
88H
01H
BR_INST_EXEC.COND Counts the number of conditional
near branch instructions executed,
but not necessarily retired.
88H
02H
BR_INST_EXEC.DIRE
CT
88H
04H
BR_INST_EXEC.INDIR Counts the number of executed
ECT_NON_CALL
indirect near branch instructions
that are not calls.
88H
07H
BR_INST_EXEC.NON
_CALLS
88H
08H
BR_INST_EXEC.RETU Counts indirect near branches that
RN_NEAR
have a return mnemonic.
88H
10H
BR_INST_EXEC.DIRE
CT_NEAR_CALL
88H
20H
BR_INST_EXEC.INDIR Counts indirect near calls, including
ECT_NEAR_CALL
both register and memory indirect,
executed.
88H
30H
BR_INST_EXEC.NEAR Counts all near call branches
_CALLS
executed, but not necessarily
retired.
88H
40H
BR_INST_EXEC.TAKE Counts taken near branches
N
executed, but not necessarily
retired.
88H
7FH
BR_INST_EXEC.ANY
Comment
Counts all unconditional near branch
instructions excluding calls and
indirect branches.
Counts all non call near branch
instructions executed, but not
necessarily retired.
Counts unconditional near call
branch instructions, excluding non
call branch, executed.
Counts all near executed branches
(not necessarily retired). This
includes only instructions and not
micro-op branches. Frequent
branching is not necessarily a major
performance issue. However
frequent branch mispredictions may
be a problem.
Vol. 3B A-83
PERFORMANCE-MONITORING EVENTS
Table A-6. Non-Architectural Performance Events In the Processor Core for Processors
Based on Intel Microarchitecture Code Name Westmere
Event
Num.
Umask
Value
Event Mask
Mnemonic
89H
01H
BR_MISP_EXEC.CON
D
Counts the number of mispredicted
conditional near branch instructions
executed, but not necessarily
retired.
89H
02H
BR_MISP_EXEC.DIRE
CT
Counts mispredicted macro
unconditional near branch
instructions, excluding calls and
indirect branches (should always be
0).
89H
04H
BR_MISP_EXEC.INDIR Counts the number of executed
ECT_NON_CALL
mispredicted indirect near branch
instructions that are not calls.
89H
07H
BR_MISP_EXEC.NON
_CALLS
89H
08H
BR_MISP_EXEC.RETU Counts mispredicted indirect
RN_NEAR
branches that have a rear return
mnemonic.
89H
10H
BR_MISP_EXEC.DIRE
CT_NEAR_CALL
89H
20H
BR_MISP_EXEC.INDIR Counts mispredicted indirect near
ECT_NEAR_CALL
calls exeucted, including both
register and memory indirect.
89H
30H
BR_MISP_EXEC.NEA
R_CALLS
89H
40H
BR_MISP_EXEC.TAKE Counts executed mispredicted near
N
branches that are taken, but not
necessarily retired.
89H
7FH
BR_MISP_EXEC.ANY
A-84 Vol. 3B
Description
Counts mispredicted non call near
branches executed, but not
necessarily retired.
Counts mispredicted non-indirect
near calls executed, (should always
be 0).
Counts all mispredicted near call
branches executed, but not
necessarily retired.
Counts the number of mispredicted
near branch instructions that were
executed, but not necessarily
retired.
Comment
PERFORMANCE-MONITORING EVENTS
Table A-6. Non-Architectural Performance Events In the Processor Core for Processors
Based on Intel Microarchitecture Code Name Westmere
Event
Num.
Umask
Value
Event Mask
Mnemonic
A2H
01H
RESOURCE_STALLS.
ANY
A2H
02H
RESOURCE_STALLS.L Counts the cycles of stall due to lack
OAD
of load buffer for load operation.
A2H
04H
RESOURCE_STALLS.R This event counts the number of
S_FULL
cycles when the number of
instructions in the pipeline waiting
for execution reaches the limit the
processor can handle. A high count
of this event indicates that there are
long latency operations in the pipe
(possibly load and store operations
that miss the L2 cache, or
instructions dependent upon
instructions further down the
pipeline that have yet to retire.
A2H
08H
RESOURCE_STALLS.S This event counts the number of
TORE
cycles that a resource related stall
will occur due to the number of
store instructions reaching the limit
of the pipeline, (i.e. all store buffers
are used). The stall ends when a
store instruction commits its data to
the cache or memory.
A2H
10H
RESOURCE_STALLS.R Counts the cycles of stall due to reOB_FULL
order buffer full.
A2H
20H
RESOURCE_STALLS.F Counts the number of cycles while
PCW
execution was stalled due to writing
the floating-point unit (FPU) control
word.
Description
Comment
Counts the number of Allocator
resource related stalls. Includes
register renaming buffer entries,
memory buffer entries. In addition
to resource related stalls, this event
counts some other events. Includes
stalls arising during branch
misprediction recovery, such as if
retirement of the mispredicted
branch is delayed and stalls arising
while store buffer is draining from
synchronizing operations.
Does not include
stalls due to
SuperQ (off core)
queue full, too
many cache
misses, etc.
When RS is full,
new instructions
can not enter the
reservation
station and start
execution.
Vol. 3B A-85
PERFORMANCE-MONITORING EVENTS
Table A-6. Non-Architectural Performance Events In the Processor Core for Processors
Based on Intel Microarchitecture Code Name Westmere
Event
Num.
Umask
Value
Event Mask
Mnemonic
A2H
40H
RESOURCE_STALLS.
MXCSR
Stalls due to the MXCSR register
rename occurring to close to a
previous MXCSR rename. The
MXCSR provides control and status
for the MMX registers.
A2H
80H
RESOURCE_STALLS.
OTHER
Counts the number of cycles while
execution was stalled due to other
resource issues.
A6H
01H
MACRO_INSTS.FUSIO Counts the number of instructions
NS_DECODED
decoded that are macro-fused but
not necessarily executed or retired.
A7H
01H
BACLEAR_FORCE_IQ
Counts number of times a BACLEAR
was forced by the Instruction
Queue. The IQ is also responsible
for providing conditional branch
prediciton direction based on a static
scheme and dynamic data provided
by the L2 Branch Prediction Unit. If
the conditional branch target is not
found in the Target Array and the IQ
predicts that the branch is taken,
then the IQ will force the Branch
Address Calculator to issue a
BACLEAR. Each BACLEAR asserted
by the BAC generates approximately
an 8 cycle bubble in the instruction
fetch pipeline.
A8H
01H
LSD.UOPS
Counts the number of micro-ops
delivered by loop stream detector.
AEH
01H
ITLB_FLUSH
Counts the number of ITLB flushes.
B0H
01H
OFFCORE_REQUEST Counts number of offcore demand
S.DEMAND.READ_DA data read requests. Does not count
TA
L2 prefetch requests.
B0H
02H
OFFCORE_REQUEST
S.DEMAND.READ_CO
DE
A-86 Vol. 3B
Description
Counts number of offcore demand
code read requests. Does not count
L2 prefetch requests.
Comment
Use cmask=1 and
invert to count
cycles
PERFORMANCE-MONITORING EVENTS
Table A-6. Non-Architectural Performance Events In the Processor Core for Processors
Based on Intel Microarchitecture Code Name Westmere
Event
Num.
Umask
Value
Event Mask
Mnemonic
B0H
04H
OFFCORE_REQUEST
S.DEMAND.RFO
Counts number of offcore demand
RFO requests. Does not count L2
prefetch requests.
B0H
08H
OFFCORE_REQUEST
S.ANY.READ
Counts number of offcore read
requests. Includes L2 prefetch
requests.
B0H
10H
OFFCORE_REQUEST
S.ANY.RFO
Counts number of offcore RFO
requests. Includes L2 prefetch
requests.
B0H
40H
OFFCORE_REQUEST
S.L1D_WRITEBACK
Counts number of L1D writebacks to
the uncore.
B0H
80H
OFFCORE_REQUEST
S.ANY
Counts all offcore requests.
B1H
01H
UOPS_EXECUTED.PO Counts number of Uops executed
RT0
that were issued on port 0. Port 0
handles integer arithmetic, SIMD and
FP add Uops.
B1H
02H
UOPS_EXECUTED.PO Counts number of Uops executed
RT1
that were issued on port 1. Port 1
handles integer arithmetic, SIMD,
integer shift, FP multiply and FP
divide Uops.
B1H
04H
UOPS_EXECUTED.PO Counts number of Uops executed
RT2_CORE
that were issued on port 2. Port 2
handles the load Uops. This is a core
count only and can not be collected
per thread.
B1H
08H
UOPS_EXECUTED.PO Counts number of Uops executed
RT3_CORE
that were issued on port 3. Port 3
handles store Uops. This is a core
count only and can not be collected
per thread.
B1H
10H
UOPS_EXECUTED.PO Counts number of Uops executed
RT4_CORE
that where issued on port 4. Port 4
handles the value to be stored for
the store Uops issued on port 3. This
is a core count only and can not be
collected per thread.
Description
Comment
Vol. 3B A-87
PERFORMANCE-MONITORING EVENTS
Table A-6. Non-Architectural Performance Events In the Processor Core for Processors
Based on Intel Microarchitecture Code Name Westmere
Event
Num.
Umask
Value
Event Mask
Mnemonic
B1H
1FH
UOPS_EXECUTED.CO Counts number of cycles there are
RE_ACTIVE_CYCLES_ one or more uops being executed
NO_PORT5
and were issued on ports 0-4. This is
a core count only and can not be
collected per thread.
B1H
20H
UOPS_EXECUTED.PO Counts number of Uops executed
RT5
that where issued on port 5.
B1H
3FH
UOPS_EXECUTED.CO Counts number of cycles there are
RE_ACTIVE_CYCLES one or more uops being executed on
any ports. This is a core count only
and can not be collected per thread.
B1H
40H
UOPS_EXECUTED.PO Counts number of Uops executed
use cmask=1,
RT015
that where issued on port 0, 1, or 5. invert=1 to count
stall cycles
B1H
80H
UOPS_EXECUTED.PO Counts number of Uops executed
RT234
that where issued on port 2, 3, or 4.
B2H
01H
OFFCORE_REQUEST
S_SQ_FULL
B3H
01H
SNOOPQ_REQUESTS Counts weighted cycles of snoopq
_OUTSTANDING.DAT requests for data. Counter 0 onl